The LLMCheck index estimates Mellum2.1-12B-A2.5B-Thinking at ~46 tok/s on an M4 at Q4_K_M, scaled from its community-measured 6-bit run on this chip with the published bandwidth model, not a measured Q4_K_M run. It needs about 10 GB of unified memory. It is JetBrains's 12.15B mixture-of-experts model (2.5B active), released under the Apache 2.0 licence.
Want to run Mellum2.1-12B-A2.5B-Thinking 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~46 tok/s |
| Provenance | estimate · how estimates are made |
| Quantization | Q4_K_M |
| Minimum RAM | 10 GB |
| Parameters | 12.15B MoE (2.5B active) |
| Chip bandwidth | 120 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 (10 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 |
|---|---|---|---|---|---|---|
| community50 tok/s | Q4_K_M | llama.cpp | 16 GB | short coding prompts | note.com | Mac mini M4, 10-core GPU. Homebrew llama.cpp llama-server, -ngl 99, -c 16384; official Q4_K_M (8.07 GB). Approximate figure, run before the Metal NaN fix (b11681).; not used to anchor estimates |
| community35.9 tok/s | 6-bit | MLX | 32 GB | 4k | omlx.ai | 10-core GPU. oMLX v0.7.1.dev1, MTP off; MLX 6-bit g64. |
Ollama — the index has not verified an Ollama tag for Mellum2.1-12B-A2.5B-Thinking, 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 →
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 | estimated154 tok/s | Q4_K_M | llama.cpp | 614 GB/s |
| M4 Pro | estimated81 tok/s | Q4_K_M | llama.cpp | 273 GB/s |
| M6 | estimated60 tok/s | Q4_K_M | llama.cpp | 170 GB/s |
JetBrains' Apache 2.0 coding-agent MoE (8 Oct 2026): 12B total, 2.5B active, a thinking model with 128K context; SWE-bench Verified 47.0 on JetBrains' Pi harness. Q4_K_M is 8 GB, so it fits a 16 GB Mac (a community run measured about 50 tok/s on a base M4 Mac mini); on Metal use llama.cpp b11681+, as older builds return NaN with k-quants. No Ollama library tag yet.
Apache 2.0 licence · 12.15B MoE · 128K context · LLMCheck capability 18/50 · see its leaderboard rank →
Mellum2.1-12B-A2.5B-Thinking needs about 10 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 Mellum2.1-12B-A2.5B-Thinking stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard