The LLMCheck index estimates Qwen3.8-Flash-Next at ~55 tok/s on an M3 Ultra at Q4_K_M, scaled from its community-measured M5 Max run with the published bandwidth model, not a measured run on this chip. It needs about 120 GB of unified memory. It is Alibaba's 125B mixture-of-experts model (6B active), released under the Qwen Custom licence.
Want to run Qwen3.8-Flash-Next faster — or step up to bigger models? Find your Mac →This is an LLMCheck index estimate from its published two-term bandwidth model, not a measured run; the index labels it so everywhere it appears. Own this Mac? Submit a real run →
| Metric | Value |
|---|---|
| Tokens per second | estimated~55 tok/s |
| Provenance | estimate · how estimates are made |
| Quantization | Q4_K_M |
| Minimum RAM | 120 GB |
| Parameters | 125B MoE (6B active) |
| Chip bandwidth | 819 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 (120 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 |
|---|---|---|---|---|---|---|
| community26.5 tok/s | 4-bit | MLX | 512 GB | — | huggingface.co | RockTalk mlx_qwen4exp custom port, plain greedy, 26.4–26.6 (llama.cpp Metal on the same machine: 26.4); 4-bit affine g64.; not used to anchor estimates |
| community25.1 tok/s | 4-bit | MLX | 512 GB | short prompt | huggingface.co | oMLX 0.6.3, mlx-community 4-bit g32; wall-clock including a short prefill, so a floor (mean of 2).; not used to anchor estimates |
Ollama — the index has not verified an Ollama tag for Qwen3.8-Flash-Next, 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: Search the Ollama library → · model card →
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
Alibaba's preview of the Qwen4 architecture (2026-08-26): 125B MoE with 6B active plus a 51B n-gram embedding, Gated DeltaNet + sparse attention, native image and video input, 262K context. SWE-Bench Pro 62.5 (vendor scaffold), LiveCodeBench v6 91.9, GPQA 91.7. 4-bit MLX is 112 GB, a 128 GB-tier model; plain-decode community runs measured 43.8–47 tok/s on an M5 Max (MTPLX 4-bit) and 57 on an M5 Ultra (llama.cpp IQ4_NL). Qwen Community 1.0 license: local use is unrestricted, but coding/office 'AI Work Assistant' products and model-as-a-service need a separate commercial license.
Qwen Custom licence · 125B MoE · 262K context · LLMCheck capability 47/50 · see its leaderboard rank →
Qwen3.8-Flash-Next needs about 120 GB of unified memory. These current Apple Silicon Macs have the headroom to run it comfortably:
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 Qwen3.8-Flash-Next stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard