Flint

Name: Flint
Author: Recursive Superintelligence

Released

2026-10-21

Last refreshed

2026-10-21

Status

Researched 1d ago

Open sourceCommercial use: permittedCodingRAGAgentsLong contextJSON / Tool use

Recursive Superintelligence releases · 1 in the last 12 monthsChangelog →

Specifications

Family: Flint
Released: 2026-10-21
Context: 256K
Max output: 64K
Parameters: 39.8B (2.5B activated)
Architecture: Mixture of Experts
Knowledge cutoff: 2026-06
Specialization: code
Openness: Open source
License: Apache 2.0OSI-approvedPermitted
Weights: Available
Code: Unknown
Training: Multi-stage

Created by

Recursive Superintelligence

Recursive Superintelligence (Recursive) is an AI lab led by CEO Richard Socher that is building the Eureka Machine, an a

San Francisco, London

Website

Pricing

Output / 1M

Free

Input / 1M

Free

Cheapest of 1 route · OpenRouter

Providers(1)

OpenRouter

Links

Website HuggingFace

About

Flint is Recursive Superintelligence's first model, released October 21, 2026 as open weights under Apache 2.0. It is a mixture-of-experts coding model with 39.8B total and about 2.5B active parameters per token, trained from scratch and built for agentic coding on local hardware: investigating codebases, fixing bugs, implementing features and writing tests, with always-on reasoning and multi-turn tool use. Its documented context configuration is 262,144 tokens (the blog says up to 512k is supported) and Recursive reports 54.9% on Terminal-Bench 2.1 and 74.2% on SWE-bench Verified.

Providers(1)

Compare API pricing across 1 tracked provider route for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Route
OpenRouter	Free	Free	Serverless

Capabilities

ReasoningJSON / Tool use

Benchmark peer barsfor Coding

SWE-bench VerifiedRank 50 of 102

Claude Fable 5

96.0

Claude Opus 5

96.0

Claude Mythos Preview

93.9

Claude Opus 4.8

88.6

Flintcurrent

74.2

LiveCodeBenchRank 33 of 75

DeepSeek V4 Pro

93.5

Gemini 3.1 Pro Preview

91.7

DeepSeek V4 Flash

91.6

Qwen3.7-Max

91.6

Flintcurrent

80.3

Benchmark scores(4)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Terminal-Bench 2.1	54.9	Terminal-Bench 2.1Observed 2026-10-21	Evaluator: Recursive (developer-reported) Notes: HF card text 'it scores 54.9% on Terminal-Bench 2.1'. Terminus 2 (XML), internet access, solutions obtained through web retrieval counted as unsuccessful. NVFP4 53.7 (chart). BF16 checkpoint; temperature 1.0, top-p 0.95, top-k 20; agentic scores average pass@1 over at least 5 attempts.	Source
SWE-bench Verified	74.2	SWE-bench VerifiedObserved 2026-10-21	Evaluator: Recursive (developer-reported) Notes: HF card text '74.2% on SWE-bench Verified'. mini-swe-agent, no internet. NVFP4 74.1 (chart). BF16 checkpoint; temperature 1.0, top-p 0.95, top-k 20; agentic scores average pass@1 over at least 5 attempts.	Source
LiveCodeBench	80.3	v6Observed 2026-10-21	Evaluator: Recursive (developer-reported) Notes: Read from first-party chart assets/flint_benchmarks_nvfp4.png ('LiveCodeBench v6', Flint BF16 80.3; NVFP4 80.8); no text source. Single-turn, average pass@1 across 5 attempts. BF16 checkpoint; temperature 1.0, top-p 0.95, top-k 20; agentic scores average pass@1 over at least 5 attempts.	Source
Instruction-Following Evaluation	95.1	IFEvalObserved 2026-10-21	Evaluator: Recursive (developer-reported) Notes: Read from first-party chart assets/flint_benchmarks_nvfp4.png ('IFEval', Flint BF16 95.1; NVFP4 94.6); no text source. Single-turn, average pass@1 across 5 attempts. BF16 checkpoint; temperature 1.0, top-p 0.95, top-k 20; agentic scores average pass@1 over at least 5 attempts.	Source