Skip to content
Weights Not Walls
Menu

Blueprint-Bench 2

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://andonlabs.com/evals/blueprint-bench-2
Score shown
Epoch's column "Score", as a percentage, higher is better. The source gives no error bars.
Results
38 in all: 2 for open-weight models, 36 for closed ones. 2 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K3, 29.5%, number 24 of 38 results.

Blueprint-Bench 2 results
#ModelWeightsSettingScoreDateSource
1gemini-4-carbon_unknownGoogle DeepMindclosedeffort unknown58.6%not datedandonlabs.com
2claude-fable-5-2_unknownAnthropicclosedeffort unknown56.6%not datedandonlabs.com
3gemini-4-argon_unknownGoogle DeepMindclosedeffort unknown54.4%not datedandonlabs.com
4claude-opus-5-5_unknownAnthropicclosedeffort unknown51.2%not datedandonlabs.com
5gpt-6-astra_unknownOpenAIclosedeffort unknown49.7%not datedandonlabs.com
6gpt-6.1-sol_unknownOpenAIclosedeffort unknown43.2%not datedandonlabs.com
7gpt-6.1-terra_unknownOpenAIclosedeffort unknown42.1%not datedandonlabs.com
8claude-fable-5-1_unknownAnthropicclosedeffort unknown41.9%not datedandonlabs.com
9claude-sonnet-5-5_unknownAnthropicclosedeffort unknown39.3%not datedandonlabs.com
10claude-fable-5_unknownAnthropicclosedeffort unknown38.6%not datedandonlabs.com
11gemini-3.8-flash_unknownGoogle DeepMindclosedeffort unknown38.6%not datedandonlabs.com
12gpt-6-sol_unknownOpenAIclosedeffort unknown36.9%not datedandonlabs.com
13gpt-5.5_unknownOpenAIclosedeffort unknown36.2%not datedandonlabs.com
14gemini-3.5-flash_unknownGoogle DeepMindclosedeffort unknown33.6%not datedandonlabs.com
15gpt-5.6-sol_unknownOpenAIclosedeffort unknown33.6%not datedandonlabs.com
16grok-4.6_unknownxAIclosedeffort unknown33.2%not datedandonlabs.com
17claude-haiku-5-5_unknownAnthropicclosedeffort unknown33.0%not datedandonlabs.com
18grok-4.7_unknownxAIclosedeffort unknown32.5%not datedandonlabs.com
19muse-wonder-1.0_unknownMeta AIclosedeffort unknown31.7%not datedandonlabs.com
20gemini-3.6-flash_unknownGoogle DeepMindclosedeffort unknown31.2%not datedandonlabs.com
21gpt-6-luna_unknownOpenAIclosedeffort unknown31.2%not datedandonlabs.com
22gpt-5.6-terra_unknownOpenAIclosedeffort unknown30.8%not datedandonlabs.com
23claude-opus-5_unknownAnthropicclosedeffort unknown30.4%not datedandonlabs.com
24moonshotai/Kimi-K3Epoch: kimi-k3_unknown, Moonshotopeneffort unknown29.5%not datedandonlabs.com
25grok-4.5_unknownxAIclosedeffort unknown27.3%not datedandonlabs.com
26gpt-5.4-2026-03-05_unknownOpenAIclosedeffort unknown27.1%not datedandonlabs.com
27gemini-3.1-pro-previewGoogle DeepMindcloseddefault26.5%not datedandonlabs.com
28claude-sonnet-5_unknownAnthropicclosedeffort unknown24.9%not datedandonlabs.com
29claude-opus-4-7_unknownAnthropicclosedeffort unknown24.5%not datedandonlabs.com
30gpt-5.6-luna_unknownOpenAIclosedeffort unknown22.6%not datedandonlabs.com
31claude-opus-4-8_unknownAnthropicclosedeffort unknown14.5%not datedandonlabs.com
32claude-sonnet-4-6_unknownAnthropicclosedeffort unknown6.7%not datedandonlabs.com
33moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshotopendefault3.9%not datedandonlabs.com
34claude-haiku-4-5-20251001_unknownAnthropicclosedeffort unknown0.0%not datedandonlabs.com
35gemini-3-flash-previewGoogle DeepMindcloseddefault0.0%not datedandonlabs.com
36gemini-robotics-er-1.6-previewGooglecloseddefault0.0%not datedandonlabs.com
37grok-4-20xAIcloseddefault0.0%not datedandonlabs.com
38grok-4.3_unknownxAIclosedeffort unknown0.0%not datedandonlabs.com

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