SWE-bench Verified

Real GitHub issue resolution on the human-verified 500-task subset of SWE-bench.

#	Model	Maker	Score	Measured	Link
1	Claude Opus 5	Anthropic	96.0 / 100	Oct 14, 2026	Source
2	Claude Sonnet 5	Anthropic	85.2 / 100	Oct 14, 2026	Source
3	Kimi K2.6	Moonshot AI	80.2 / 100	Oct 14, 2026	Source
4	Mistral Medium 3.5	Mistral AI	77.6 / 100	Apr 29, 2026	Source
5	Qwen3.5-397B-A17B	Alibaba Qwen	76.2 / 100	Oct 14, 2026	Source
6	Gemini 3.1 Pro Preview	Google DeepMind	75.6 / 100	Oct 14, 2026	Source
7	GPT-5	OpenAI	73.6 / 100	Oct 14, 2026	Source
8	DeepSeek V4 Pro (0813)	DeepSeek	73.6 / 100	Oct 14, 2026	Source
9	Claude Haiku 4.5	Anthropic	73.3 / 100	Oct 14, 2026	Source
10	Nemotron 3 Ultra 550B-A55B	NVIDIA	71.9 / 100	Oct 14, 2026	Source
11	Qwen3-Coder-480B-A35B-Instruct	Alibaba Qwen	69.6 / 100	Oct 14, 2026	Source
12	MAI-Code-1-Flash	Microsoft AI	51.6 / 100	Oct 14, 2026	Source
13	Command A+	Cohere	51.0 / 100	Oct 14, 2026	Source

Scores are the benchmark owners’ published results, recorded with the date we verified them. This benchmark feeds the League Score — see methodology.