MiniMax-M3.1-Flash runs at 25 tok/s (25.3 as recorded) on a 256 GB M3 Ultra, a community run reported by omlx.ai (MLX, oQ3e, 1k context). It needs about 189 GB of unified memory. It is MiniMax's 428B mixture-of-experts model (23B active), released under the MiniMax Community licence.
llama.cpp needs b11861+ (PR #31097 fixed garbled output; Homebrew's 0.7.0 is older), Ollama is cloud-only, released mlx-lm is text-only; the NVFP4 originals need oMLX 0.7.3.dev1 or mlx-lm main (#2003).
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 | community25 tok/s |
| Provenance | community run · omlx.ai |
| Quantization | oQ3e |
| Engine | MLX |
| Context | 1k |
| Figure dated | 2026-10-24 |
| Minimum RAM | 189 GB |
| Parameters | 428B MoE (23B 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 (189 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 |
|---|---|---|---|---|---|---|
| community25.3 tok/s | oQ3e | MLX | 256 GB | 1k | omlx.ai | 60-core GPU (819 GB/s, the same as the 80-core part). oMLX v0.7.2; mlx-community MiniMax-M3.1-Flash-oQ3e-mtp with MTP/DSpark, DFlash and expert offload off; 182.1 GB peak. 18.0 tok/s at 4k, 17.3 at 8k. |
Ollama — the index has not verified an Ollama tag for MiniMax-M3.1-Flash, 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 → · model card →
The 25 tok/s figure above came from MLX at oQ3e; 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 Ultra | community35 tok/s | oQ3e | MLX | 1228 GB/s |
MiniMax's 428B MoE (23B active), weights published 23 Oct 2026: text, image and video in, 1M context, shipped quantization-aware as ~250 GB of NVFP4/MXFP8 weights with a DSpark draft head. The MiniMax Community License is non-commercial by default: commercial use needs a 'Built with MiniMax M3.1-Flash' credit and notice, plus written approval above $20M revenue. Card figures are agentic, e.g. DeepSWE v1.1 73.7 (mini-SWE-agent). 3-bit MLX builds (187–196 GB) need a 256 GB Mac Studio in practice; community oMLX runs measured 34.5 tok/s on a 256 GB M5 Ultra and 25.3 on a 256 GB M3 Ultra (oQ3e, MTP off). llama.cpp handles GGUFs of the official checkpoint from b11861 (issue #31063 fixed 24 Oct).
MiniMax Community licence · 428B MoE · 1000K context · LLMCheck capability 48/50 · see its leaderboard rank →
MiniMax-M3.1-Flash needs about 189 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 MiniMax-M3.1-Flash stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard