The LLMCheck index estimates Qwen3.8-27B at ~21 tok/s on an M1 Max at Q4_K_M, scaled from its community-measured 8-bit run on this chip with the published bandwidth model, not a measured Q4_K_M run. It needs about 19 GB of unified memory. It is Alibaba's 27.8B dense model, released under the Apache 2.0 licence.
Want to run Qwen3.8-27B 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. The bandwidth formula’s typical error for dense models is ±9%, against 75 measured runs of 20 models (how it is computed). Own this Mac? Submit a real run →
| Metric | Value |
|---|---|
| Tokens per second | estimated~21 tok/s |
| Provenance | estimate · how estimates are made |
| Quantization | Q4_K_M |
| Minimum RAM | 19 GB |
| Parameters | 27.8B |
| Chip bandwidth | 400 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 (19 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 |
|---|---|---|---|---|---|---|
| community11.38 tok/s | 8-bit | LM Studio | 64 GB | 41-token chat prompt, 541 generated, thinking on (medium) | github.com | MacBook Pro, M1 Max (10-core CPU, 32-core GPU), 64 GB, macOS 27.0.1, plugged in. LM Studio 0.4.24 (lms chat --stats), MLX runtime 1.13.1; Qwen3.8-27B-MLX-8bit (the submission does not name the uploader); single run, 30,000-token context allocated, reasoning effort medium. Submitted as GitHub issue #1 on 8 Oct 2026. |
Ollama — install it (brew install ollama or the app from ollama.com), then pull the model by its tag, checked on ollama.com on 2026-11-14:
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Ollama page → · Hugging Face 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 Ultra | sourced48 tok/s | oQ4e | MLX | 1228 GB/s |
| M3 Ultra | community38 tok/s | 4-bit | MLX | 819 GB/s |
| M5 Max | community31 tok/s | 4-bit | MLX | 614 GB/s |
| M4 Max | estimated26 tok/s | Q4_K_M | MLX | 546 GB/s |
| M3 Max | estimated19 tok/s | Q4_K_M | MLX | 400 GB/s |
| M5 Pro | sourced17 tok/s | 4-bit | llama.cpp | 307 GB/s |
| M4 Pro | estimated13 tok/s | Q4_K_M | Ollama | 273 GB/s |
| M6 | community10 tok/s | 4-bit | MLX | 170 GB/s |
Alibaba's Apache 2.0 27.8B dense vision-language model (2026-08-14): image and video input, 262K context (1M with YaRN), thinking mode with reasoning_effort. Runs 4-bit in ~19 GB, so it fits a 24 GB Mac; ~29 tok/s estimated on M5 Max, and BGR's review measured ~55 on an M5 Ultra and ~40 on an M3 Ultra — within 4% of the index's estimates. Model card: SWE-Bench Pro 61.7 (vendor scaffold), Terminal-Bench 2.1 73.0, LiveCodeBench v6 90.3, GPQA 89.2. 1,000+ GGUF quants; Ollama, llama.cpp and LM Studio supported.
Apache 2.0 licence · 27.8B · 262K context · LLMCheck capability 46/50 · see its leaderboard rank →
Estimated, not measured: the index’s minimum for Qwen3.8-27B (19 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. hybrid: full attention on 16 of 64 layers; Gated DeltaNet layers keep a constant state. A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M1 Max Mac on macOS 15, the lower default; macOS 26 allows more): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).
| Context | KV cache (16-bit) | RAM needed | RAM, q8 KV | Smallest tier that fits | Within the default GPU limit |
|---|---|---|---|---|---|
| 8K | ~0.5 GB | ~19.3 GB | ~19.1 GB | 24 GB Mac raise GPU limit | 32 GB Mac |
| 32K | ~2.1 GB | ~20.9 GB | ~19.9 GB | 32 GB Mac | |
| 64K (coding-agent context) | ~4.3 GB | ~23.0 GB | ~21.0 GB | 32 GB Mac raise GPU limit (macOS 15) | 36 GB Mac |
| 128K | ~8.6 GB | ~27.3 GB | ~23.2 GB | 32 GB Mac raise GPU limit | 48 GB Mac |
Qwen3.8-27B needs about 19 GB of unified memory. These Apple Silicon Macs have enough to run it. Apple builds the higher-memory configurations to order; Amazon does not stock them.
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 Qwen3.8-27B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard