FrontierCode

Measures: An agent working in a real open-source repository must produce a pull request its maintainers would actually merge; graded by an ensemble of unit tests, rubric review and scope/quality verifiers, with the headline Main score the Mean@5 over the 100 hardest of 150 private tasks, and runs that consult solution-bearing sources zeroed.

Maintainer: Cognition
Introduced: 2026
Status: Active
Saturation: Headroom
Typical provenance: Independent
In scoring set: Yes

Official leaderboard

Leaderboard

Show all results

Independent and third-party results only.

Leaderboard
Model	Main score (Mean@5)	Runs flagged for unfair internet use	Provenance	Source	Evaluated
Mistral Medium 3.5 Mistral AI	8.0%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Qwen3.7-Plus Alibaba Group	10.2%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Inkling Thinking Machines Lab	14.0%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
MiniMax-M3 MiniMax	14.7%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
DeepSeek-V4-Pro DeepSeek · None effort	17.6%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
DeepSeek-V4-Flash DeepSeek · Unknown effort	18.8%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Sonnet 4.6 Anthropic · Unknown effort	24.3%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GLM-5.2 Z.ai (Zhipu AI) · None effort	24.5%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Opus 4.6 Anthropic · Unknown effort	26.6%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-5.4 mini OpenAI · Unknown effort	27.0%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
DeepSeek-V4-Pro DeepSeek · High effort	28.6%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Kimi K2.7 Code Moonshot AI	30.1%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GLM-5.3-Flash Z.ai (Zhipu AI) · Max effort	31.8%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Gemini 3.6 Flash Google DeepMind · Unknown effort	34.4%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Opus 4.7 Anthropic · Unknown effort	38.5%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-5.6 Luna OpenAI · Unknown effort	39.8%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GLM-5.3 Z.ai (Zhipu AI) · Max effort	40.1%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Gemini 3.8 Flash Google DeepMind · Medium effort	41.2%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-5.6 Terra OpenAI · Unknown effort	41.3%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-6 Luna OpenAI · Max effort	42.4%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Grok 4.5 SpaceXAI (xAI) · Unknown effort	42.4%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Sonnet 5 Anthropic · Unknown effort	42.7%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-5.5 OpenAI · Unknown effort	43.0%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Gemini 3.7 Flash Google DeepMind · Unknown effort	43.6%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Kimi K3 Moonshot AI · Unknown effort	44.2%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Opus 4.8 Anthropic · Unknown effort	46.5%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-5.6 Sol OpenAI · Unknown effort	47.5%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Grok 4.6 SpaceXAI (xAI) · Unknown effort	48.0%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-6 Sol OpenAI · Max effort	49.3%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Fable 5.1 Anthropic · Medium effort	50.9%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Sonnet 5.5 Anthropic · Xhigh effort	52.1%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
GPT-6 Astra OpenAI · Max effort	53.3%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Opus 5 Anthropic · Max effort	53.4%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—
Claude Fable 5 Anthropic · Unknown effort	53.5%	—	Third party	https://cognition.com/frontiercode via Epoch AI	—

Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0.

Sources

Epoch AI