MiniCPM5-2B runs at 144 tok/s (144.2 as recorded) on a 24 GB M5 Pro, a community run reported by omlx.ai (MLX, 4-bit, 1k context). It needs about 3 GB of unified memory. It is OpenBMB's 2.52B dense model, released under the Apache 2.0 licence.
Want to run MiniCPM5-2B faster — or step up to bigger models? Find your Mac →This is a community run: a real run, reported by its source and linked below. The index records it as given and never adjusts it.
| Metric | Value |
|---|---|
| Tokens per second | community144 tok/s |
| Provenance | community run · omlx.ai |
| Quantization | 4-bit |
| Engine | MLX |
| Time to first token | 0.215s |
| Prompt processing (prefill) | community4554 tok/s at 8K context |
| Context | 1k |
| Figure dated | 2026-11-04 |
| Minimum RAM | 3 GB |
| Parameters | 2.52B |
| Chip bandwidth | 307 GB/s |
Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (3 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.
| Speed | Quant | Engine | Mac RAM | Context | Source | Note |
|---|---|---|---|---|---|---|
| sourced147.34 tok/s | 4-bit | MLX | — | decode at 4K context depth | huggingface.co | Competitor baseline on SparkLLM's Spark-X2.6-1.7B card chart 'Mac M5 Pro · 16-core' (the 16-core-GPU M5 Pro has the same 307 GB/s): stock MLX-LM 0.32.2 on an INT4 mlx-community checkpoint. 81.47 tok/s at 32K. |
| community144.2 tok/s | 4-bit | MLX | 24 GB | 1k | omlx.ai | 20-core GPU. oMLX v0.7.3rc1, MTP/DFlash off; openbmb MiniCPM5-2B-MLX (4-bit); 2.2 GB peak. 136.2 tok/s at 4k. |
Measured figures only, each from its source’s own session; the index does not estimate long-context or prefill speed, so a missing cell is —. Each @ column is plain decode with that many tokens already in the prompt, from the same session as the row’s short-context figure: decode slows as the KV cache it reads grows. Prefill is prompt processing, the tokens per second a runtime reads a prompt at; with long agent contexts it sets how long you wait before the first token.
| Run | Decode (row) | @4K | @8K | @32K | @128K | Prefill | TTFT |
|---|---|---|---|---|---|---|---|
| MLX · 4-bit · huggingface.co | sourced147.34 tok/s | sourced147.34 tok/s | — | sourced81.47 tok/s | sourced31.75 tok/s | sourced4071 tok/s | — |
| MLX · 4-bit · 24 GB Mac · omlx.ai | community144.2 tok/s | community136.2 tok/s | community118.7 tok/s | — | — | community4554 tok/s | community0.215 s |
Ollama — the index has not verified an Ollama tag for MiniCPM5-2B, so it doesn’t print one (a guessed tag fails with “pull model manifest: file does not exist”). Search the Ollama library →
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Hugging Face repo → · Hugging Face model card →
The 144 tok/s figure above came from MLX at 4-bit; other runtimes and quants run at different speeds.
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
| Chip | Speed | Quant | Engine | Bandwidth |
|---|---|---|---|---|
| M5 Max | estimated225 tok/s | Q4_K_M | llama.cpp | 614 GB/s |
OpenBMB's Apache 2.0 2.5B dense model (6 Sep 2026), 128K context. Artificial Analysis ranks it among the two strongest open models of 4B or fewer parameters. Vendor card: GPQA Diamond 70.2, MMLU-Pro 70.8. Q4_K_M is 1.6 GB; stock llama.cpp and the official MLX 4-bit build run it, and SparkLLM's card chart lists 147 tok/s on an M5 Pro (mlx-lm 0.32.2, 4-bit).
Apache 2.0 licence · 2.52B · 128K context · LLMCheck capability 21/50 · see its leaderboard rank →
Estimated, not measured: the index’s minimum for MiniCPM5-2B (3 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. GQA on all 42 layers (2 kv heads x 128). A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M5 Pro Mac, which runs macOS 26 or later): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).
MiniCPM5-2B needs about 3 GB of unified memory. These Apple Silicon Macs have enough to run it.
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.
See how MiniCPM5-2B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard