MiMo-V2.6-Flash runs at 59 tok/s on a 256 GB M3 Ultra, a community run reported by github.com (MLX, 4-bit, 4K context). It needs about 170 GB of unified memory. It is Xiaomi's 309B mixture-of-experts model (15B active), released under the MIT licence.
Want to run MiMo-V2.6-Flash 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 | community59 tok/s |
| Provenance | community run · github.com |
| Quantization | 4-bit |
| Engine | MLX |
| Context | 4K |
| Figure dated | 2026-09 |
| Minimum RAM | 170 GB |
| Parameters | 309B MoE (15B 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 (170 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 |
|---|---|---|---|---|---|---|
| community59.4 tok/s | 4-bit | MLX | 256 GB | 128-token serial decode | huggingface.co | RL checkpoint (same architecture as MOPD). oMLX 0.7.0.dev2 / MLX 0.32.2 / mlx-lm 0.31.3, MTP not executed; 4-bit affine dense + MXFP4 experts. |
| community59 tok/s | 4-bit | MLX | 256 GB | 4K | github.com | rapid-mlx 0.15.0, no MTP or speculative execution; mimo-v2.6-flash-4bit (171.7 GB). 49 tok/s at 30K. |
| community55.1 tok/s | 4-bit | MLX | 512 GB | 1k | omlx.ai | 80-core GPU. oMLX v0.7.0, MTP/DFlash off; MiMo-V2.6-Flash-RL-MLX-4bit-MTP (the RL checkpoint, same architecture as MOPD); 154.3 GB peak. 53.7 tok/s at 4k in the same session (twfrx909). |
Ollama — the index has not verified an Ollama tag for MiMo-V2.6-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 59 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 Ultra | community62 tok/s | 4-bit | MLX | 1228 GB/s |
Xiaomi's MIT-licensed 309B MoE with 15B active: text, image, video and audio in, 1M context. The 27 Sep 2026 MOPD checkpoint supersedes the RL release (Terminal-Bench 2.1 87.6 and DeepSWE v1.1 67.9 are RL-checkpoint card figures). The smallest GGUF is 126 GB and 4-bit MLX builds are ~172 GB, so it starts at a 192 GB Mac Studio. A community oMLX 4-bit run measured 61.7 tok/s on an M5 Ultra (256 GB, MTP off), and community MLX runs on an M3 Ultra measured 55–59; all used the RL checkpoint, which has the same architecture as the MOPD checkpoint listed here. The index's earlier M5 Ultra estimate was ~79.
MIT licence · 309B MoE · 1000K context · LLMCheck capability 49/50 · see its leaderboard rank →
MiMo-V2.6-Flash needs about 170 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 MiMo-V2.6-Flash stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard