Maple Preview 20B-A1B runs at 169 tok/s on a 16 GB M4, a published figure reported by github.com (MLX, Ternary). It needs about 6 GB of unified memory. It is DeepGrove's 20B mixture-of-experts model (1B active), released under the MIT licence.
Runs only through DeepGrove's own forks (mlx-lm-deepgrove, deepgrove-ai/llama.cpp); its cards name no stock llama.cpp, mlx-lm, LM Studio or Ollama route.
This is a published figure: 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 | sourced169 tok/s |
| Provenance | published figure · github.com |
| Quantization | Ternary |
| Engine | MLX |
| Figure dated | 2026-08 |
| Minimum RAM | 6 GB |
| Parameters | 20B MoE (1B 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 (6 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 |
|---|---|---|---|---|---|---|
| sourced169 tok/s | Ternary | MLX | 16 GB | — | github.com | DeepGrove's mlx-lm fork (mlx-lm-deepgrove) README, exact output head; its approximate --flash-head mode reaches 218, the figure behind the model card's '200+ tokens/sec on a Mac mini M4'. The README says the M4 figures come from an earlier version of the fork and does not state the M4 run's prompt length. |
Ollama — the index has not verified an Ollama tag for Maple Preview 20B-A1B, 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 169 tok/s figure above came from MLX at Ternary; 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 Pro | sourced387 tok/s | Ternary | MLX | 307 GB/s |
DeepGrove's MIT-licensed 20B MoE with about 1.5B active and ternary weights (about 5 GB). DeepGrove reports 387 tok/s on an M5 Pro and 169 on a Mac mini M4 in its own mlx-lm fork, with the exact output head (its approximate 'flash head' mode reaches 477 and 218). It runs only through DeepGrove's forks of mlx-lm and llama.cpp. Ternary decoding is compute-bound, so the bandwidth formula does not describe it and the index quotes only its measured figures.
MIT licence · 20B MoE · 131K context · LLMCheck capability 20/50 · see its leaderboard rank →
Maple Preview 20B-A1B needs about 6 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 Maple Preview 20B-A1B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard