unsloth/flint-GGUF · Hugging Face

Source: https://huggingface.co/unsloth/flint-GGUF

https://huggingface.co/unsloth/flint-GGUF#read-our-how-to-run-flint-guideRead our How toRun Flint Guide!

See further below for quant benchmarks.See below for Flint running in Unsloth Desktop with thinking controls:

https://huggingface.co/unsloth/flint-GGUF#flint-unsloth-gguf-benchmarksFlint Unsloth GGUF Benchmarks

Recursive Flint

Blog|Hugging Face|Local engine|Documentation

Flint is Recursive’s mixture-of-experts language model for coding agents, with 39.8 billion total parameters and approximately 2.5 billion active per token. It is designed to investigate codebases, fix bugs, implement features, and write tests.

Flint supports text input and output, reasoning, and multi-turn tool use. In our release evaluations, it scores 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2, and 74.2% on SWE-bench Verified.

This repository contains the BF16 checkpoint, model configuration, tokenizer, chat template, and an SGLang serving kit. Quantized variants are prepared separately for NVIDIA and Apple Silicon runtimes.

https://huggingface.co/unsloth/flint-GGUF#evaluationEvaluation

Flint results use temperature 1.0, top-p 0.95, and top-k 20. Terminal-Bench uses Terminus 2; SWE-bench uses mini-swe-agent. Agentic scores report average pass@1 over at least five attempts. † marks scores measured by Recursive; other comparison scores are developer-reported.

Evaluation setups vary across sources. Flint’s Terminal-Bench score counts solutions obtained through web retrieval as unsuccessful; developer-reported scores may not apply the same adjustment.

https://huggingface.co/unsloth/flint-GGUF#model-specificationsModel specifications

PropertyFlintDeveloperRecursiveArchitectureDecoder-only mixture-of-experts TransformerTotal parameters39.8BActive parameters per tokenApproximately 2.5B, including the output head¹Layers49: 2 dense layers followed by 47 MoE layersAttention35 sliding-window layers and 14 full-attention layersSliding window513 tokens: the current token and 512 preceding tokensAttention heads32 query heads and 4 key/value heads; head dimension 128Attention normalizationPer-head QK RMSNorm and learned per-head attention sinksPosition encodingRoPE on sliding-window layers; no positional encoding on full-attention layersExperts per MoE layer256 routed experts, with 6 selected per token, plus an always-active shared expertDocumented context configuration262,144 tokens, including prompt and generated outputVocabulary150,144 tokensInput / outputText / textStated knowledge cutoffJune 2026 ¹ The active-parameter convention includes the output head and excludes the full input embedding matrix. The active body alone is approximately 2.2B parameters. “40B-A2.5B” is the rounded model-size designation.

https://huggingface.co/unsloth/flint-GGUF#recommended-generation-settingsRecommended generation settings

SettingValueTemperature1.0Top-p / top-k0.95 / 20Output budget for hard problemsAt least 32768 tokens, including reasoning The server applies these sampling settings when a request omits them. Flint supports OpenAI-compatible tool calls. Use the checkpoint’s chat template and preserve returnedreasoning\_contentin conversation history. Prompt and output share the context window.

https://huggingface.co/unsloth/flint-GGUF#trainingTraining

Flint was trained from scratch through pretraining, mid-training, supervised fine-tuning, and reinforcement learning. Training includes text and code, instruction following, and agentic tasks.

https://huggingface.co/unsloth/flint-GGUF#limitations-and-responsible-useLimitations and responsible use

Flint’s supervised fine-tuning included safety-oriented examples, but the model can still produce incorrect or harmful content, faulty code, or inappropriate tool calls. Test generated code and review consequential changes before applying them. Tool-using applications should enforce their own permissions and execution boundaries; structured tool-call grammar does not prevent the model from choosing an unsafe or incorrect action.

Long reasoning can increase latency and exhaust the output budget before an answer is complete. The context configuration is a capacity limit, not a guarantee of uniform accuracy across all sequence lengths or tasks. Results can vary with checkpoint precision and inference runtime; evaluate the specific checkpoint and runtime you intend to deploy.

https://huggingface.co/unsloth/flint-GGUF#license-and-reportingLicense and reporting

Flint is released under theApache 2.0 license. Hosted API terms and data practices are separate from the weight license.

For model and serving issues, contact[email protected]and include the checkpoint revision, runtime version, hardware, launch command, and a minimal reproduction. Report safety concerns or suspected misuse to[email protected], and security vulnerabilities to[email protected].