Local AI Advisor

Can MacBook Pro M5 32 GB run Recursive Flint (40B-A2.5B)?

32 GB unified · 153 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: 40–65 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 WebUIExcellentMXFP440–65 tok/sunder 1 s8KDetails
Coding questionswith ContinueBorderlineMXFP440–65 tok/s≈3 s16KDetails
Coding in a repositorywith AiderBorderlineMXFP430–50 tok/s≈8 s32KDetails
Agentic codingwith OpenCodeBorderlineMXFP430–50 tok/s≈10 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP420–35 tok/s≈30 s64KDetails

Speeds are bandwidth-based estimates; no direct benchmark exists for this pair. “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.