{
 "version": "2026-10-05",
 "count": 244,
 "methodology": "LLMCheck index — normalized estimates and sourced benchmarks for local LLMs on Apple Silicon. See https://llmcheck.net/methodology. MoE rows (refit 5 Oct 2026): small MoEs (<= 40B total) use the same two-term model on ACTIVE-parameter bytes with a 5.16 ms per-token overhead, fitted to six fully specified plain-decode measurements on three chips (mean absolute error ~12%); large MoEs (> 40B) keep 0.42 x the dense formula on active bytes, fitted to the only fully specified large-MoE runs (DeepSeek V4 Flash, 2-bit). A model with its own measured rows has its other chips scaled from the nearest-bandwidth measured row. Estimated rows carry no time-to-first-token: the estimator models decode only, so TTFT appears only where a source published it.",
 "reference_hardware": "M5 Max 128GB Unified Memory",
 "benchmarks": [
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 168,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 86,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "MLX",
   "tps": 34,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 8,
   "engine": "Ollama",
   "tps": 34,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 24,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 161,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "Ollama",
   "tps": 39,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 8,
   "engine": "Ollama",
   "tps": 33,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 82,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E2B",
   "params": "2.3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 261,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E2B",
   "params": "2.3B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 147,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E2B",
   "params": "2.3B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 62,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E2B",
   "params": "2.3B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 8,
   "engine": "Ollama",
   "tps": 44,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 161,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M5 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 91,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 82,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 33,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 8,
   "engine": "Ollama",
   "tps": 23,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 106,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B",
   "quant": "Q4_K_M",
   "chip": "M5 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 72,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 100,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 67,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 31B",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 26,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 31B",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 31B",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 161,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 8,
   "engine": "Ollama",
   "tps": 23,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M3 Pro",
   "ram": 18,
   "engine": "MLX",
   "tps": 48,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 82,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "LM Studio",
   "tps": 18,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 15,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 10,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "Ollama",
   "tps": 91,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 8B",
   "params": "8B",
   "quant": "Q8_0",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 24,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 16,
   "engine": "Ollama",
   "tps": 17,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 91,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 20,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 16,
   "engine": "Ollama",
   "tps": 17,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 11,
   "ttft": null,
   "date": "2025-11",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 102,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 50,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 19,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 13,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 91,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "Ollama",
   "tps": 20,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M3 Pro",
   "ram": 18,
   "engine": "Ollama",
   "tps": 25,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 11,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 91,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "LM Studio",
   "tps": 17,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 64,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 11,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 16,
   "engine": "Ollama",
   "tps": 11,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 55,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "LM Studio",
   "tps": 26,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 10,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 55,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 16,
   "engine": "MLX",
   "tps": 10,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 55,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 26,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 30,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "LM Studio",
   "tps": 14,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 27,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 30B-A3B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 116,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 30B-A3B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 76,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 30B-A3B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 36,
   "engine": "Ollama",
   "tps": 94,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 35B-A3B",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 117,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 35B-A3B",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "Ollama",
   "tps": 111,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "Ollama",
   "tps": 25,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 32,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 25,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "LM Studio",
   "tps": 23,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 32B",
   "params": "32B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 36,
   "engine": "Ollama",
   "tps": 17,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.3 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 12,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 2.5 72B",
   "params": "72B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 120B",
   "params": "117B-A5.1B",
   "quant": "MXFP4",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "Ollama",
   "tps": 58,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 4 Scout",
   "params": "109B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 19,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 59,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral 7B",
   "params": "7B",
   "quant": "Q8_0",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 28,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 47,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 52,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 14B",
   "params": "14B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 31,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 102,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 Mini",
   "params": "3.8B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 153,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 4B",
   "params": "4B",
   "quant": "Q8_0",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 47,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 35B-A3B",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 96,
   "engine": "Ollama",
   "tps": 96,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q8_0",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 52,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 44,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 16,
   "engine": "Ollama",
   "tps": 11,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 20,
   "ttft": null,
   "date": "2026-02",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Ministral 3 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "LM Studio",
   "tps": 10,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 96,
   "engine": "MLX",
   "tps": 20,
   "ttft": null,
   "date": "2026-01",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 8,
   "engine": "Ollama",
   "tps": 17,
   "ttft": null,
   "date": "2025-12",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 39,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3 4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 161,
   "ttft": null,
   "date": "2026-03",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 120,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 86,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 4",
   "params": "119B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 46,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-235B-A22B",
   "params": "235B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 15,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron-Cascade 2",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 117,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron-Cascade 2",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 111,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron-Cascade 2",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 77,
   "ttft": null,
   "date": "2026-04",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 3.2 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 33,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 3.2 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 3.2 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 32,
   "engine": "Ollama",
   "tps": 15,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Hermes 4 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 12,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Hermes 4 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 11,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "SmolLM3 3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 199,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "SmolLM3 3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 106,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "SmolLM3 3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M3",
   "ram": 16,
   "engine": "Ollama",
   "tps": 43,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "SmolLM3 3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 8,
   "engine": "Ollama",
   "tps": 43,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "SmolLM3 3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M1",
   "ram": 8,
   "engine": "Ollama",
   "tps": 30,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Devstral Small 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 33,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Devstral Small 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Devstral Small 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 32,
   "engine": "Ollama",
   "tps": 15,
   "ttft": null,
   "date": "2026-05",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.5-Air",
   "params": "106B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 27,
   "ttft": null,
   "date": "2026-07",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.5-Air",
   "params": "106B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 24,
   "ttft": null,
   "date": "2026-07",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash",
   "params": "284B-A13B",
   "quant": "Q2_K",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 39,
   "ttft": 0.4,
   "date": "2026-08",
   "provenance": "community",
   "source": "https://modelfit.io/blog/deepseek-v4-flash-0731-128gb-macbook/"
  },
  {
   "model": "DeepSeek V4 Flash",
   "params": "284B-A13B",
   "quant": "Q2_K",
   "chip": "M3 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 27,
   "ttft": 0.6,
   "date": "2026-08",
   "provenance": "community",
   "source": "https://modelfit.io/blog/deepseek-v4-flash-0731-128gb-macbook/"
  },
  {
   "model": "GLM 5.2",
   "params": "753B-A40B",
   "quant": "IQ1_S",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "LM Studio",
   "tps": 22,
   "ttft": 1.5,
   "date": "2026-07",
   "provenance": "community",
   "source": "https://codersera.com/blog/how-to-run-glm-5-2-locally-2026/"
  },
  {
   "model": "GLM 5.2",
   "params": "753B-A40B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 512,
   "engine": "MLX",
   "tps": 15,
   "ttft": 2.0,
   "date": "2026-07",
   "provenance": "community",
   "source": "https://kingy.ai/blog/glm-5-2-256gb-mac-studio-local/"
  },
  {
   "model": "Muse Glimmer 30B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "LM Studio",
   "tps": 10,
   "ttft": 0.8,
   "date": "2026-08",
   "provenance": "community",
   "source": "https://kingy.ai/blog/muse-glimmer-private-offline-ai-agent/"
  },
  {
   "model": "Muse Glimmer 30B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 27,
   "ttft": 0.4,
   "date": "2026-08",
   "provenance": "sourced",
   "source": "https://huggingface.co/blog/muse-glimmer"
  },
  {
   "model": "Muse Glimmer 30B",
   "params": "30B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 24,
   "ttft": 0.5,
   "date": "2026-08",
   "provenance": "sourced",
   "source": "https://huggingface.co/blog/muse-glimmer"
  },
  {
   "model": "LFM2.5-2.6B",
   "params": "2.6B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 220,
   "ttft": 0.1,
   "date": "2026-08",
   "provenance": "sourced",
   "source": "https://www.liquid.ai/blog/lfm2-5-2-6b"
  },
  {
   "model": "Maple Preview 20B-A1B",
   "params": "20B-A1B",
   "quant": "Ternary",
   "chip": "M5 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 281,
   "ttft": 0.1,
   "date": "2026-08",
   "provenance": "sourced",
   "source": "https://huggingface.co/deepgrove/maple-preview"
  },
  {
   "model": "Maple Preview 20B-A1B",
   "params": "20B-A1B",
   "quant": "Ternary",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 200,
   "ttft": 0.1,
   "date": "2026-08",
   "provenance": "sourced",
   "source": "https://huggingface.co/deepgrove/maple-preview"
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 27,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 64,
   "engine": "Ollama",
   "tps": 20,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M5 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 15,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 14,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "KAT-Coder-V2.5",
   "params": "35B-A3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 117,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "KAT-Coder-V2.5",
   "params": "35B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 111,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "KAT-Coder-V2.5",
   "params": "35B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 77,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron 3.5 Lightning",
   "params": "30B-A3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 103,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron 3.5 Lightning",
   "params": "30B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 97,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nemotron 3.5 Lightning",
   "params": "30B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 64,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Laguna S 2.1",
   "params": "118B-A8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 38,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Laguna S 2.1",
   "params": "118B-A8B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 35,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Laguna XS 2.1",
   "params": "33B-A3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 117,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Laguna XS 2.1",
   "params": "33B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 77,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Inkling-Small",
   "params": "276B-A12B",
   "quant": "Q2_K",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 42,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Bonsai 27B",
   "params": "27B",
   "quant": "1-bit",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Bonsai 27B",
   "params": "27B",
   "quant": "1-bit",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 6,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Bonsai 27B",
   "params": "27B",
   "quant": "1-bit",
   "chip": "M2",
   "ram": 8,
   "engine": "MLX",
   "tps": 5,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nanbeige4.2-3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 199,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nanbeige4.2-3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "Ollama",
   "tps": 51,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Nanbeige4.2-3B",
   "params": "3B",
   "quant": "Q4_K_M",
   "chip": "M2",
   "ram": 8,
   "engine": "Ollama",
   "tps": 43,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Apertus 1.5 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 91,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Apertus 1.5 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "Ollama",
   "tps": 20,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Apertus 1.5 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 12,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.7-Flash",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 26,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.7-Flash",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 48,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.7-Flash",
   "params": "31B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "Ollama",
   "tps": 12,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.3 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M1 Ultra",
   "ram": 128,
   "engine": "Ollama",
   "tps": 15,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M1 Ultra",
   "ram": 128,
   "engine": "Ollama",
   "tps": 15,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B-A3B",
   "quant": "Q4_K_M",
   "chip": "M1 Ultra",
   "ram": 64,
   "engine": "MLX",
   "tps": 138,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M1 Ultra",
   "ram": 64,
   "engine": "MLX",
   "tps": 38,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B-A4B",
   "quant": "Q4_K_M",
   "chip": "M1 Ultra",
   "ram": 64,
   "engine": "Ollama",
   "tps": 119,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 120B",
   "params": "117B-A5.1B",
   "quant": "MXFP4",
   "chip": "M1 Ultra",
   "ram": 128,
   "engine": "Ollama",
   "tps": 72,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.3 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 96,
   "engine": "Ollama",
   "tps": 8,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B-A3B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 105,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 32,
   "engine": "MLX",
   "tps": 20,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 3 27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 32,
   "engine": "Ollama",
   "tps": 20,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 32,
   "engine": "Ollama",
   "tps": 62,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 3.2 24B",
   "params": "24B",
   "quant": "Q4_K_M",
   "chip": "M2 Max",
   "ram": 32,
   "engine": "Ollama",
   "tps": 22,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.3 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "Ollama",
   "tps": 16,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash",
   "params": "284B-A13B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 32,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Inkling-Small",
   "params": "276B-A12B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 512,
   "engine": "MLX",
   "tps": 35,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-235B-A22B",
   "params": "235B-A22B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 20,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Hunyuan Hy3",
   "params": "295B-A21B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "LM Studio",
   "tps": 21,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-Coder-Next",
   "params": "80B-A3B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 84,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-Coder-Next",
   "params": "80B-A3B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 77,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 24,
   "engine": "MLX",
   "tps": 8,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-27B",
   "params": "27B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 24,
   "engine": "MLX",
   "tps": 9,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "KAT-Coder-V2.5",
   "params": "35B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 32,
   "engine": "MLX",
   "tps": 56,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 8B",
   "params": "8B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 16,
   "engine": "MLX",
   "tps": 28,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 9B",
   "params": "9B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 16,
   "engine": "MLX",
   "tps": 25,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Phi-4 14B",
   "params": "14B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 16,
   "engine": "MLX",
   "tps": 16,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 E4B",
   "params": "4B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 16,
   "engine": "MLX",
   "tps": 54,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "LFM2.5-2.6B",
   "params": "2.6B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 16,
   "engine": "MLX",
   "tps": 79,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash",
   "params": "284B-A13B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 45,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Inkling-Small",
   "params": "276B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 49,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-235B-A22B",
   "params": "235B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 29,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Mistral Small 4",
   "params": "119B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 79,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 120B",
   "params": "117B-A5.1B",
   "quant": "MXFP4",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 97,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.3 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek R1 70B",
   "params": "70B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3-Coder-Next",
   "params": "80B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 130,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-4.5-Air",
   "params": "106B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 49,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Laguna S 2.1",
   "params": "118B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 68,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Llama 3.1 405B",
   "params": "405B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 512,
   "engine": "MLX",
   "tps": 4,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 2.5 72B",
   "params": "72B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 96,
   "engine": "MLX",
   "tps": 23,
   "ttft": null,
   "date": "2026-08",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3-Flash",
   "params": "320B-A18B",
   "quant": "IQ3_XXS",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 27,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3-Flash",
   "params": "320B-A18B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 24,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3-Flash",
   "params": "320B-A18B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 35,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-Flash-Next",
   "params": "125B-A6B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 49,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-Flash-Next",
   "params": "125B-A6B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 44,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-Flash-Next",
   "params": "125B-A6B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 62,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-Flash-Next",
   "params": "125B-A6B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 84,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash Vision-Exp",
   "params": "305B-A13B",
   "quant": "Q2_K",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "llama.cpp",
   "tps": 39,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash Vision-Exp",
   "params": "305B-A13B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "llama.cpp",
   "tps": 32,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "DeepSeek V4 Flash Vision-Exp",
   "params": "305B-A13B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "llama.cpp",
   "tps": 46,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 29,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 26,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 19,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 48,
   "engine": "Ollama",
   "tps": 13,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3",
   "params": "753B-A40B",
   "quant": "Q2_K",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 27,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3",
   "params": "753B-A40B",
   "quant": "Q2_K",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 19,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3",
   "params": "753B-A40B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 512,
   "engine": "MLX",
   "tps": 17,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GLM-5.3",
   "params": "753B-A40B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 512,
   "engine": "MLX",
   "tps": 11,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "LM Studio",
   "tps": 55,
   "ttft": 0.5,
   "date": "2026-09",
   "provenance": "sourced",
   "source": "https://www.bgr.com/2264114/mac-studio-m5-ultra-review/"
  },
  {
   "model": "Qwen3.8-27B",
   "params": "27.8B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 96,
   "engine": "LM Studio",
   "tps": 40,
   "ttft": null,
   "date": "2026-09",
   "provenance": "sourced",
   "source": "https://www.bgr.com/2264114/mac-studio-m5-ultra-review/"
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "params": "122B-A10B",
   "quant": "unstated",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "LM Studio",
   "tps": 80,
   "ttft": null,
   "date": "2026-09",
   "provenance": "sourced",
   "source": "https://www.bgr.com/2264114/mac-studio-m5-ultra-review/"
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "params": "122B-A10B",
   "quant": "unstated",
   "chip": "M3 Ultra",
   "ram": 96,
   "engine": "LM Studio",
   "tps": 60,
   "ttft": null,
   "date": "2026-09",
   "provenance": "sourced",
   "source": "https://www.bgr.com/2264114/mac-studio-m5-ultra-review/"
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "params": "122B-A10B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 47,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "params": "122B-A10B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 42,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.5 122B-A10B",
   "params": "122B-A10B",
   "quant": "Q4_K_M",
   "chip": "M3 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 32,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "MiMo-V2.6-Flash",
   "params": "309B-A15B",
   "quant": "MXFP4",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "MiMo-V2.6-Flash",
   "params": "309B-A15B",
   "quant": "MXFP4",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 42,
   "ttft": null,
   "date": "2026-09",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B-A3B",
   "quant": "4-bit",
   "chip": "M5 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 126,
   "ttft": null,
   "date": "2026-08",
   "provenance": "community",
   "source": "https://omlx.ai/benchmarks/performance/nx08c47t"
  },
  {
   "model": "Qwen 3.6-35B-A3B",
   "params": "35B-A3B",
   "quant": "4-bit",
   "chip": "M6",
   "ram": 32,
   "engine": "MLX",
   "tps": 64,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Nemotron 3.5 Lightning",
   "params": "30B-A3B",
   "quant": "Q4_K_M",
   "chip": "M6",
   "ram": 32,
   "engine": "llama.cpp",
   "tps": 45,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Gemma 4 26B-A4B",
   "params": "26B-A4B",
   "quant": "QAT",
   "chip": "M6",
   "ram": 32,
   "engine": "LM Studio",
   "tps": 48,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Gemma 4 31B",
   "params": "31B",
   "quant": "QAT",
   "chip": "M6",
   "ram": 32,
   "engine": "LM Studio",
   "tps": 8,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Muse Glimmer 30B",
   "params": "30B",
   "quant": "unstated",
   "chip": "M6",
   "ram": 32,
   "engine": "LM Studio",
   "tps": 9,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "GPT-oss 20B",
   "params": "21B-A3.6B",
   "quant": "unstated",
   "chip": "M6",
   "ram": 32,
   "engine": "LM Studio",
   "tps": 45,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Gemma 4 12B",
   "params": "12B",
   "quant": "QAT",
   "chip": "M6",
   "ram": 32,
   "engine": "LM Studio",
   "tps": 19,
   "ttft": null,
   "date": "2026-10",
   "provenance": "community",
   "source": "https://github.com/HDZucht/Mac-Mini-M6-LLM-benchmarks"
  },
  {
   "model": "Qwen 3 30B-A3B",
   "params": "30B-A3B",
   "quant": "4-bit",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "MLX",
   "tps": 110,
   "ttft": null,
   "date": "2026-01",
   "provenance": "sourced",
   "source": "https://arxiv.org/abs/2601.19139"
  },
  {
   "model": "Qwen 3 30B-A3B",
   "params": "30B-A3B",
   "quant": "4-bit",
   "chip": "M4 Max",
   "ram": 128,
   "engine": "llama.cpp",
   "tps": 90,
   "ttft": null,
   "date": "2026-01",
   "provenance": "sourced",
   "source": "https://arxiv.org/abs/2601.19139"
  },
  {
   "model": "GPT-oss 20B",
   "params": "21B-A3.6B",
   "quant": "MXFP4",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 103,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 20B",
   "params": "21B-A3.6B",
   "quant": "MXFP4",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 97,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 20B",
   "params": "21B-A3.6B",
   "quant": "MXFP4",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 64,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "GPT-oss 20B",
   "params": "21B-A3.6B",
   "quant": "MXFP4",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 34,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M5 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 64,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M4 Max",
   "ram": 64,
   "engine": "MLX",
   "tps": 57,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M4 Pro",
   "ram": 24,
   "engine": "MLX",
   "tps": 30,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "Gemma 4 12B",
   "params": "12B",
   "quant": "Q4_K_M",
   "chip": "M4",
   "ram": 16,
   "engine": "MLX",
   "tps": 13,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "MiniMax M2.5",
   "params": "230B-A10B",
   "quant": "Q4_K_M",
   "chip": "M5 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 57,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  },
  {
   "model": "MiniMax M2.5",
   "params": "230B-A10B",
   "quant": "Q4_K_M",
   "chip": "M3 Ultra",
   "ram": 256,
   "engine": "MLX",
   "tps": 41,
   "ttft": null,
   "date": "2026-10",
   "provenance": "estimated",
   "source": null
  }
 ],
 "provenance_policy": "Each row carries provenance: 'sourced' (a published figure from an accountable source — a vendor's model card or announcement, or an independent press review that documents its setup — source URL attached), 'community' (one person's run: a blog post, forum thread or benchmark submission — source URL attached), or 'estimated' (derived from the LLMCheck estimation model — memory-bandwidth scaling + quantization math; see https://llmcheck.net/methodology#estimation). No figure is claimed as first-party measured."
}