The best local LLM for a Mac Studio M3 Ultra (512 GB) is Qwen3.8-Flash-Next at 63 tok/s (estimated). With 512 GB of unified memory it runs 69 of the models in the LLMCheck index — from compact options up to 753B-class models. For everyday chat and coding, Qwen3.8-Flash-Next is the sweet spot. Full ranking below.
Considering this Mac?
Mac Studio M3 Ultra (512GB) on Amazon → · not sure? the Mac Advisor picks a config →
Ranked by LLMCheck suitability (capability balanced against speed on the M3 Ultra). Click a model for its full benchmark and setup. Every speed carries its provenance mark: most are index estimates from the memory-bandwidth model, and a filled square is a published measurement — submit a real run →
| # | Model | Size | License | Speed | Capability |
|---|---|---|---|---|---|
| 1 | Qwen3.8-Flash-Next | 125B | Qwen Custom | estimated63 tok/s | 47/50 |
| 2 | MiMo-V2.6-Flash | 309B | MIT | estimated30 tok/s | 49/50 |
| 3 | GLM 5.2 | 753B | MIT | community15 tok/s | 50/50 |
| 4 | GLM-5.3 | 753B | GLM-5.3 | estimated12 tok/s | 50/50 |
| 5 | GLM-5.3-Flash | 320B | MIT | estimated25 tok/s | 48/50 |
| 6 | DeepSeek V4 Flash Vision-Exp | 305B | MIT | estimated33 tok/s | 47/50 |
| 7 | Qwen3.8-27B | 27.8B | Apache 2.0 | sourced40 tok/s | 46/50 |
| 8 | DeepSeek V4 Flash | 284B | MIT | estimated30 tok/s | 47/50 |
| 9 | Inkling-Small | 276B | Apache 2.0 | estimated30 tok/s | 46/50 |
| 10 | Qwen 3.6-27B | 27B | Apache 2.0 | estimated40 tok/s | 44/50 |
| 11 | Qwen3-235B-A22B | 235B | Apache 2.0 | estimated19 tok/s | 46/50 |
| 12 | KAT-Coder-V2.5 | 35B | Apache 2.0 | estimated69 tok/s | 39/50 |
Showing the top 12 of 69 models that fit in 512 GB. See the full leaderboard or all benchmarks.
The fastest way to get started is Ollama. Install it, then find the top pick for your Mac in its library and copy the exact tag it lists:
Find Qwen3.8-Flash-Next in the Ollama library → · model card on Hugging Face →
Prefer a GUI? LM Studio gives you a one-click download and chat window. For step-by-step help see our Ollama install guide, or open the Qwen3.8-Flash-Next on M3 Ultra benchmark page for exact settings.
The Mac Studio M3 Ultra (512 GB) comfortably runs 69 of the models in the LLMCheck index, led by Qwen3.8-Flash-Next. Grab one and start running LLMs offline today:
As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence our rankings.
Qwen3.8-Flash-Next (125B, Qwen Custom) is the best all-round pick at 63 tok/s (estimated) on the M3 Ultra. If you want maximum speed, Maple Preview 20B-A1B hits 463 tok/s (estimated); for maximum capability, GLM 5.2 still fits in 512 GB.
About 69 of the 89 models in the LLMCheck leaderboard fit in 512 GB of unified memory, from compact models up to GLM 5.2 (753B).
Yes. A 70B model in Q4 quantization needs roughly 40–44 GB of memory, which fits in 512 GB with headroom for context.
512 GB is plenty for local AI — you can run capable 30B–70B-class models. Because Apple Silicon uses unified memory, that figure is both your system RAM and your VRAM.