Local AI Advisor

Can Laptop · GeForce RTX 5090 Laptop · 32 GB RAM run Recursive Flint (40B-A2.5B)?

24 GB VRAM + 32 GB RAM · 896 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 64K-token documents.

Best quantization for chat: MXFP4 (20 GB of weights) on Ollama. Generation: 240–390 tok/s, leaving about 2.1 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 WebUIExcellentMXFP4240–390 tok/sunder 1 s8KDetails
Coding questionswith ContinueExcellentMXFP4220–370 tok/sunder 1 s16KDetails
Coding in a repositorywith AiderComfortableMXFP4180–310 tok/s≈2 s32KDetails
Agentic codingwith OpenCodeComfortableMXFP4170–280 tok/s≈2 s32KDetails
Long documents (64K)with Open WebUIBorderlineMXFP4120–190 tok/s≈7 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.