Skip to content
Weights Not Walls
Menu

GSO-Bench

Reported

No description yet: we only describe an eval set when we can do it correctly.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://gso-bench.github.io/index.html
Score shown
Epoch's column "Score OPT@1", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
Results
42 in all: 3 for open-weight models, 39 for closed ones. 3 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K2-Instruct, 4.9%, number 31 of 42 results.

GSO-Bench results
#ModelWeightsSettingScoreDateSource
1claude-fable-5-1_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands88.2%27 Sep 2026gso-bench.github.io
2gpt-6-astra_unknownOpenAIclosedeffort unknown; Scaffold: OpenHands79.4%27 Sep 2026gso-bench.github.io
3claude-fable-5_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands78.4%27 Sep 2026gso-bench.github.io
4gpt-5.6-sol_unknownOpenAIclosedeffort unknown; Scaffold: OpenHands76.5%27 Sep 2026gso-bench.github.io
5claude-opus-4-8_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands47.1%12 Jul 2026gso-bench.github.io
6claude-opus-4-7AnthropicclosedScaffold: OpenHands44.1%27 Apr 2026GSO Leaderboard, via Epoch AI
7claude-opus-4-7_highAnthropicclosedeffort high; Scaffold: OpenHands44.1%27 Apr 2026gso-bench.github.io
8claude-opus-4-6_highAnthropicclosedeffort high; Scaffold: OpenHands41.2%27 Apr 2026GSO Leaderboard, via Epoch AI
9gpt-5.5_xhighOpenAIclosedeffort xhigh; Scaffold: OpenHands40.2%27 Apr 2026gso-bench.github.io
10claude-sonnet-5_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands37.3%12 Jul 2026gso-bench.github.io
11claude-opus-4-6AnthropicclosedScaffold: OpenHands33.3%27 Apr 2026GSO Leaderboard, via Epoch AI
12claude-opus-4-6_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands33.3%27 Apr 2026gso-bench.github.io
13gpt-5.4-2026-03-05_xhighOpenAIclosedeffort xhigh; Scaffold: OpenHands31.4%10 Mar 2026gso-bench.github.io
14gpt-5.2-2025-12-11_highOpenAIclosedeffort high; Scaffold: OpenHands27.5%18 Dec 2025gso-bench.github.io
15claude-opus-4-5-20251101AnthropicclosedScaffold: OpenHands26.5%not datedGSO Leaderboard
16claude-opus-4-5-20251101_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands26.5%28 Nov 2025gso-bench.github.io
17gpt-5.4-2026-03-05_highOpenAIclosedeffort high; Scaffold: OpenHands25.5%10 Mar 2026gso-bench.github.io
18gemini-3.1-pro-previewGoogle DeepMindclosedScaffold: OpenHands22.6%9 Mar 2026gso-bench.github.io
19gemini-3-pro-previewGoogle DeepMindclosedScaffold: OpenHands18.6%28 Nov 2025gso-bench.github.io
20claude-sonnet-4-5-20250929_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands14.7%10 Oct 2025gso-bench.github.io
21claude-sonnet-4-5-20250929AnthropicclosedScaffold: OpenHands14.7%not datedGSO Leaderboard
22gpt-5.1-2025-11-13_unknownOpenAIclosedeffort unknown; Scaffold: OpenHands13.7%18 Dec 2025gso-bench.github.io
23gpt-5.1-2025-11-13_highOpenAIclosedeffort high; Scaffold: OpenHands13.7%not datedGSO Leaderboard, via Epoch AI
24gemini-3-flash-previewGoogle DeepMindclosedScaffold: OpenHands9.8%19 Dec 2025gso-bench.github.io
25o3-2025-04-16_highOpenAIclosedeffort high; Scaffold: OpenHands8.8%9 Jun 2025gso-bench.github.io
26claude-opus-4-20250514AnthropicclosedScaffold: OpenHands6.9%not datedGSO Leaderboard
27claude-opus-4-20250514_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands6.9%10 Aug 2025gso-bench.github.io
28gpt-5-2025-08-07_highOpenAIclosedeffort high; Scaffold: OpenHands6.9%10 Aug 2025gso-bench.github.io
29claude-sonnet-4-20250514AnthropicclosedScaffold: OpenHands4.9%not datedGSO Leaderboard
30claude-sonnet-4-20250514_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands4.9%30 May 2025gso-bench.github.io
31moonshotai/Kimi-K2-InstructEpoch: Kimi-K2-Instruct, MoonshotopenScaffold: OpenHands4.9%26 Aug 2025gso-bench.github.io
32Qwen/Qwen3-Coder-480B-A35B-InstructEpoch: Qwen3-Coder-480B-A35B-Instruct, AlibabaopenScaffold: OpenHands4.9%2 Sep 2025gso-bench.github.io
33claude-3-5-sonnet-20241022AnthropicclosedScaffold: OpenHands4.6%30 May 2025gso-bench.github.io
34gemini-2.5-proGoogle DeepMindclosedScaffold: OpenHands3.9%10 Oct 2025gso-bench.github.io
35gemini-2.5-pro-preview-06-05Google DeepMindclosedScaffold: OpenHands3.9%10 Oct 2025GSO Leaderboard, via Epoch AI
36claude-3-7-sonnet-20250219AnthropicclosedScaffold: OpenHands3.8%not datedGSO Leaderboard
37claude-3-7-sonnet-20250219_unknownAnthropicclosedeffort unknown; Scaffold: OpenHands3.8%30 May 2025gso-bench.github.io
38o4-mini-2025-04-16_highOpenAIclosedeffort high; Scaffold: OpenHands3.6%30 May 2025gso-bench.github.io
39zai-org/GLM-4.5-AirEpoch: GLM-4.5-Air, Z.ai (Zhipu AI),Tsinghua UniversityopenScaffold: OpenHands2.9%26 Aug 2025gso-bench.github.io
40o3-mini-2025-01-31_highOpenAIclosedeffort high; Scaffold: OpenHands1.3%30 May 2025gso-bench.github.io
41o3-mini-2025-01-31_lowOpenAIclosedeffort low; Scaffold: OpenHands1.3%30 May 2025GSO Leaderboard, via Epoch AI
42gpt-4o-2024-11-20OpenAIclosedScaffold: OpenHands0.0%30 May 2025gso-bench.github.io

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method