Local AI Advisor

Can Mac Studio M2 Max 32 GB run Recursive Flint (40B-A2.5B)?

32 GB unified · 400 GB/s memory bandwidth · Recursive Flint (40B-A2.5B) has 39.8B parameters (2.5B active per token).

Short answer

Yes, excellent for chat; borderline for coding questions.

Best quantization for chat: MXFP4 (20 GB of weights) on LM Studio. Generation: 130–170 tok/s, leaving about 3.5 GB free for your other apps.

How it rates for each workload

Same model and machine, different jobs. Agents and long documents send far bigger prompts than chat, so they need much more speed and memory.

WorkloadVerdictQuantGenerationFirst replyContext
Chatwith Open WebUIExcellentMXFP4130–170 tok/sunder 1 s8KDetails
Coding questionswith ContinueBorderlineMXFP4120–160 tok/s≈3 s16KDetails
Coding in a repositorywith AiderBorderlineMXFP4100–130 tok/s≈9 s32KDetails
Agentic codingwith OpenCodeBorderlineMXFP490–120 tok/s≈10 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP465–85 tok/s≈35 s64KDetails

Speeds are estimates calibrated against measured benchmarks on similar hardware. “First reply” is the time to read the workload’s opening prompt. How we calculate

Different tool, context size or workload?

These pages use typical settings. The full check models your exact setup.