- input
- $0.00
- output
- $0.00
- Context
- 262Ktokens
- Listed
- 3w ago
- Size
- 40Bparameters
Cost/1M tokens
Recursive's Flint is a mixture-of-experts language model built from scratch by a small team with help from the company's Eureka Machine, its automated AI research system. The launch announcement headlines it as a fast, intelligent model for coding locally on a laptop, claiming it runs more than five times faster than dense models of similar total size with comparable coding performance. It is built for fast agentic coding on local hardware, so users can keep their data on their own machine.
- Language model
- Open weights · 40B
- Reasoning
- Tools
How it compares
Additional source evidence
- Source model
- recursive/flint
- Source family
- flint
- Last updated
- 2026-10-21
- Knowledge cutoff
- 2026-06
- Model context
- 262K
- Model output limit
- 262K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- Yes
- Temperature control
- Yes
Downloadable weights
Official references
Complete source record (JSON)
{
"id": "recursive/flint",
"name": "Flint",
"description": "Open-weight mixture-of-experts model with about 2.5B active parameters for local coding agents",
"family": "flint",
"last_updated": "2026-10-21",
"knowledge": "2026-06",
"attachment": false,
"reasoning": true,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 262144,
"output": 262144
},
"release_date": "2026-10-21",
"open_weights": true,
"license": "Apache-2.0",
"links": [
{
"url": "https://huggingface.co/RecursiveSI/flint",
"label": "Model card",
"type": "model_card"
},
{
"url": "https://www.recursive.com/updates/introducing-recursive-flint",
"label": "Introducing Recursive Flint",
"type": "announcement"
}
],
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/RecursiveSI/flint"
},
{
"label": "Hugging Face (INT4)",
"url": "https://huggingface.co/RecursiveSI/flint-int4",
"quantization": "int4"
}
]
}Frequently asked questions
Running locally
At Q4 Flint needs an estimated 30.4 GB of VRAM, at Q8 about 49.4 GB, and at FP16 about 92.5 GB. All figures are estimates: weights plus about 15% and 0.5 GB for a short context.
No. Based on estimates, Flint does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16.
Yes. The Q4 estimate is 30.4 GB of VRAM, and assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits is 48 GB.
The smallest published build at 4-bit or higher is flint:40b-a2b-q4_1 (Q4_1, 26 GB download), with an estimated 30.4 GB of VRAM.
Run ollama run flint:latest, a 23 GB download. There are also 4 other local Ollama builds.
The only runtime with recorded support is Ollama. Support for llama.cpp, vLLM, LM Studio or MLX is not recorded.
The default Ollama build flint:latest is 23 GB. Across the 5 builds with a published size, the smallest is 23 GB and the largest is 80 GB.
Privacy and data
Yes, Flint has open weights, so you can self-host it, for example with ollama run flint:latest.
Capabilities and specs
The context window is 262K tokens (262,144) and the maximum output is 236K tokens (235,929).
No, Flint only accepts text input.
Yes, tool calling (function calling) is supported.
Yes, structured output (JSON schema or JSON mode) is supported.
Yes, a reasoning mode is supported and it is always on.
The knowledge cutoff of Flint is 2026-06.
Quality and benchmarks
Closest alternatives, matched on model kind and release date with no comparable price, are LightOnOCR-3 0.8B by Lightonai, MiniCPM-V 4.7 35B-A3B by OpenBMB and MiniCPM-V 4.7 1B by OpenBMB.
LightOnOCR-3 0.8BFlint vs LightOnOCR-3 0.8BMiniCPM-V 4.7 35B-A3BFlint vs MiniCPM-V 4.7 35B-A3BMiniCPM-V 4.7 1BFlint vs MiniCPM-V 4.7 1B
Licensing and access
Flint has open weights, released under the Apache-2.0 license.
Yes, the Apache 2.0 license is a permissive license that allows commercial use.
The weights are available on the Hugging Face repository RecursiveSI/flint. There are also 5 Ollama builds.
Basics
Flint was made by Recursive and was first listed on OpenRouter on October 19, 2026.
Yes, Flint has open weights.
Generated from Vertical Tabs data, updated November 12, 2026.







Discussion
No comments yet
Be the first to share your thoughts.