Decompose into subgoals
Splits the problem into named subgoals before solving any of them. Helps on long chains, adds latency on short ones.
Prompt
// system
List the subgoals required to solve the problem, then solve each in order, then state the final answer.
// user
{{problem}}
Score history
Model versionVersionScore · CIHoldoutState
gemini-3.2-prov181.0 73.2–86.979.6fresh
llama-4.2-405bv171.4 63.0–78.670.4fresh
claude-opus-4.8v183.3 75.9–88.881.5fresh
gpt-5.6-solv184.1 76.8–89.583.3fresh
mistral-large-3v176.2 68.0–82.875.9fresh
claude-fable-5v185.7 78.5–90.885.2fresh
current scorev1
81.0
95% CI 73.2 to 86.9 · n=126
Holdout score79.6
Judgeprogrammatic
Cost per run$0.843
Median latency2.8s
LicenseCC-BY-4.0
Author@ilya
0 endorsements
Sign in to endorse this entry.
Across models
gemini-3.2-pro81.0fresh
llama-4.2-405b71.4fresh
claude-opus-4.883.3fresh
gpt-5.6-sol84.1fresh
mistral-large-376.2fresh
claude-fable-585.7fresh
Versions
v1First submission81.0