Skip to content
Weights Not Walls
Menu

PostTrainBench

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.
Score shown
Epoch's column "Average (%)", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
Results
22 in all: 4 for open-weight models, 18 for closed ones. 4 linked to a model page here.

Best open-weight result: zai-org/GLM-5.3-Flash, 36.8%, number 9 of 22 results.

PostTrainBench results
#ModelWeightsSettingScoreDateSource
1claude-fable-5-1_maxAnthropicclosedeffort max; Scaffold: Claude Code44.6%not datedvia Epoch AI
2claude-fable-5-2_maxAnthropicclosedeffort max; Scaffold: Claude Code44.6%not datedvia Epoch AI
3claude-opus-5-5_maxAnthropicclosedeffort max; Scaffold: Claude Code43.8%not datedvia Epoch AI
4gpt-6-astra_maxOpenAIclosedeffort max; Scaffold: Codex CLI41.9%not datedvia Epoch AI
5gemini-4-carbon_highGoogle DeepMindclosedeffort high; Scaffold: OpenCode41.0%not datedvia Epoch AI
6gemini-4-argon_highGoogle DeepMindclosedeffort high; Scaffold: OpenCode40.4%not datedvia Epoch AI
7gpt-6.1-sol_maxOpenAIclosedeffort max; Scaffold: Codex CLI38.8%not datedvia Epoch AI
8claude-fable-5_maxAnthropicclosedeffort max; Scaffold: Claude Code37.3%not datedvia Epoch AI
9zai-org/GLM-5.3-FlashEpoch: glm-5.3-flash_max, Z.ai (Zhipu AI)openeffort max; Scaffold: Claude Code36.8%not datedvia Epoch AI
10claude-opus-5_unknownAnthropicclosedeffort unknown; Scaffold: Claude Code35.7%not datedvia Epoch AI
11zai-org/GLM-5.3Epoch: glm-5.3_max, Z.ai (Zhipu AI)openeffort max; Scaffold: Claude Code34.5%not datedvia Epoch AI
12moonshotai/Kimi-K3Epoch: kimi-k3_unknown, Moonshotopeneffort unknown; Scaffold: Claude Code31.5%not datedvia Epoch AI
13gpt-5.6-sol_maxOpenAIclosedeffort max; Scaffold: Codex CLI31.0%not datedvia Epoch AI
14claude-opus-4-8_maxAnthropicclosedeffort max; Scaffold: Claude Code31.0%not datedvia Epoch AI
15claude-opus-4-8_highAnthropicclosedeffort high; Scaffold: Claude Code30.7%not datedvia Epoch AI
16zai-org/GLM-5.2Epoch: glm-5.2_max, Z.ai (Zhipu AI)openeffort max; Scaffold: Claude Code30.4%not datedvia Epoch AI
17claude-opus-4-7_xhighAnthropicclosedeffort xhigh; Scaffold: Claude Code24.3%not datedvia Epoch AI
18muse-wonder-1.0_xhighMeta AIclosedeffort xhigh; Scaffold: OpenCode24.2%not datedvia Epoch AI
19gpt-5.5_xhighOpenAIclosedeffort xhigh; Scaffold: Codex CLI23.9%not datedvia Epoch AI
20grok-4.5_highxAIclosedeffort high; Scaffold: Cursor CLI22.6%not datedvia Epoch AI
21gemini-3.1-pro-previewGoogle DeepMindclosedScaffold: OpenCode18.1%not datedvia Epoch AI
22gpt-5.4-2026-03-05_highOpenAIclosedeffort high; Scaffold: Codex CLI17.8%not datedvia Epoch AI

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