GSO-Bench

Software optimisation tasks measured against expert speed-ups.

Published by

GSO-Benchdata via Epoch AI Benchmarking Hub

Category

Coding

Index weight

0.5

Models

31

Data as of

9 Sept 2026

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

1. 1Claude Opus 4.847.1%
2. 2Claude Opus 4.744.1%
3. 3Claude Opus 4.744.1%
4. 4Claude Opus 4.641.2%
5. 5GPT-5.540.2%
6. 6Claude Sonnet 537.3%
7. 7Claude Opus 4.633.3%
8. 8GPT-5.431.4%
9. 9GPT-5.227.4%
10. 10Claude Opus 4.526.5%
11. 11GPT-5.425.5%
12. 12Gemini 3.1 Pro22.6%
13. 13Gemini 3 Pro18.6%
14. 14Claude Sonnet 4.514.7%
15. 15GPT-5.113.7%

SearchAll providersAlibabaAnthropicGoogleMoonshot AIOpenAIZhipu AIOpen weights31 of 31

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Opus 4.8Anthropic	47.1%	–	64.7	–
2	Claude Opus 4.7Anthropic	44.1%	–	65.5	–
3	Claude Opus 4.7Anthropic	44.1%	–	61.3	–
4	Claude Opus 4.6Anthropic	41.2%	–	61.3	–
5	GPT-5.5OpenAI	40.2%	–	63.7	–
6	Claude Sonnet 5Anthropic	37.3%	–	61.2	–
7	Claude Opus 4.6Anthropic	33.3%	–	63.1	–
8	GPT-5.4OpenAI	31.4%	–	61.3	–
9	GPT-5.2OpenAI	27.4%	–	55.5	–
10	Claude Opus 4.5Anthropic	26.5%	–	58.0	–
11	GPT-5.4OpenAI	25.5%	–	58.8	–
12	Gemini 3.1 ProGoogle	22.6%	–	63.9	–
13	Gemini 3 ProGoogle	18.6%	–	61.6	–
14	Claude Sonnet 4.5Anthropic	14.7%	–	52.3	–
15	GPT-5.1OpenAI	13.7%	–	59.4	–
16	GPT-5.1OpenAI	13.7%	–	53.8	–
17	Gemini 3 FlashGoogle	9.8%	–	56.6	–
18	o3OpenAI	8.8%	–	52.9	–
19	Claude Opus 4Anthropic	6.9%	–	50.7	–
20	GPT-5OpenAI	6.9%	–	54.5	–
21	Claude Sonnet 4Anthropic	4.9%	–	49.5	–
22	Kimi K2Moonshot AIopen	4.9%	–	48.8	–
23	Qwen3 Coder 480BAlibabaopen	4.9%	–	48.5	–
24	Claude 3.5 SonnetAnthropic	4.6%	–	45.1	–
25	Gemini 2.5 ProGoogle	3.9%	–	54.1	–
26	Claude 3.7 SonnetAnthropic	3.8%	–	49.8	–
27	o4-miniOpenAI	3.6%	–	48.0	–
28	GLM-4.5-AirZhipu AIopen	2.9%	–	46.0	–
29	o3-miniOpenAI	1.3%	–	47.8	–
30	o3-miniOpenAI	1.3%	–	40.4	–
31	GPT-4oOpenAI	0.0%	–	42.3	–