Interactive Program Induction - Rebuilt From One Seed - Rebuilt From One Seed: leaderboard

Pinned appendix A4.T3; same released instruction-tuned models; exact subset/query/prior protocol in caption. A4T3 three seeds, evaluator rebuiltfromsinglevisibleexample. A5T7 100problems, temp1, three weightingrules. A5T8 100problems, nativeprompttemperatures andfixedselectionrule. A5T9 112sampledprograms, nooraclequeries. No cost/query-count metrics or duplicate Ours columns. Metric: Program accuracy (0–1). Source: https://arxiv.org/html/2611.18485v1

Metric: Program accuracy (0–1). Source: arxiv.org. Saturation forecast: Around January 2028. 6 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Qwen 2.5 32B Instruct0.28#1257
2Qwen 2.5 7B Instruct0.231#2706
3OLMo 3 7B Instruct0.224#6916
4Llama 3.1 8B Instruct0.205#4576
5Ministral-8B-Instruct-24100.165#4569
6Llama 3.2 3B Instruct0.157#7230

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=interactive-program-induction-rebuilt-from-one-seed-rebuilt-from-one-seed · How It Works · Data refreshed daily, snapshot 2026-11-17.