Recursive Flint 40B A2.5B VRAM requirements
Recursive · 39.8B params (2.5B active) · Apache 2.0. Estimates, not vendor numbers.
Recursive Flint 40B A2.5B (39.8B parameters, 2.5B active per token) needs about 25 GB of VRAM or unified memory at Q4, 30 GB at Q5 and 44 GB at Q8 (weights plus about 1.5 GB headroom, our estimate). The smallest common setup that runs it is 32GB GPU (RTX 5090).
| Quant | Download | Memory needed |
|---|---|---|
| Q4 | 24 GB | 25 GB |
| Q5 | 28 GB | 30 GB |
| Q8 | 42 GB | 44 GB |
Will it run on my machine?
| Setup | Runs? | Speed est. |
|---|---|---|
| 8GB GPU (RTX 4060) | No | n/a |
| 12GB GPU (RTX 3060 / 4070) | No | n/a |
| 16GB GPU (RTX 4060 Ti 16GB) | No | n/a |
| 24GB GPU (RTX 4090) | No | n/a |
| 32GB GPU (RTX 5090) | Yes, Q5 | ~604 tok/s |
| Mac 16GB (M4) | No | n/a |
| Mac 48GB (M4 Pro) | Yes, Q5 | ~92 tok/s |
| Mac 128GB (M4 Max) | Yes, Q8 | ~123 tok/s |
| PC, 32GB RAM, no GPU | No | n/a |
Best quant that fits each setup, with estimated generation speed from memory bandwidth. Real speed depends on the runtime, context length and settings.
Mixture of experts: 39.8B total, about 2.5B active per token, 262K context. Coding-agent model released Oct 21, 2026. Recursive's 4-bit builds are 22.6 to 26 GB and it recommends 48 GB of memory.
Out of plan usage? Recursive Flint, a free local coding model, vs a paid plan.
FAQ
How much VRAM does Recursive Flint 40B A2.5B need?
About 25 GB at Q4, 30 GB at Q5 and 44 GB at Q8, including about 1.5 GB headroom. These are estimates from the parameter count on the official model card.
Can Recursive Flint 40B A2.5B run on 8GB of VRAM?
Not fully on the GPU. At Q4 it needs about 25 GB; partial CPU offload works but is much slower.
Can Recursive Flint 40B A2.5B run on a Mac?
Yes. Mac 48GB (M4 Pro), Mac 128GB (M4 Max) can run it.
Which quant of Recursive Flint 40B A2.5B should I use?
Use the highest quant that fits with room to spare: Q8 is near full quality, Q5 is a good balance, and Q4 is the smallest that usually keeps quality acceptable.
Why is Recursive Flint 40B A2.5B fast for its size?
It is a mixture of experts: 39.8B parameters in total but only 2.5B are used per token, so it reads less memory per token. You still need memory for all 39.8B.
Similar-size models
Qwen3.8 27B (27.8B) · Muse Glimmer 30B (29.6B) · Qwen3 Coder 30B A3B (30.5B) · Gemma 4 31B (30.7B) · DeepSeek R1 Distill Qwen 32B (32.8B) · Qwen3.6 35B A3B (35B)