PolicyBench - US Exact Match - tanf: leaderboard

The tanf output of PolicyBench US, without tools. Read countries.us.heatmap exact, with the native common per-output denominator and symmetric reference exclusions. Aggregate programStats rows across models are not model scores. Metric: Exact match (%). Source: https://policybench.org

Metric: Exact match (%). Source: policybench.org. Saturation forecast: Estimated already saturated. 52 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)99#99
2Claude Sonnet 4.699#173
3GPT-5.599#50
4Claude Haiku 4.599#580
5DeepSeek V4 Pro99#198
6Claude Opus 4.799#82
7GPT-5.6 Sol99#33
8GPT-5.4 Mini99#401
9Claude Opus 4.899#62
10Gemini 3 Flash (Preview)99#151
11MiniMax-M399#278
12Gemini 3.8 Flash99#39
13Claude Opus 599#23
14Gemini 3.5 Flash99#107
15GPT-6 Astra99#8

No result here: #2 GPT-6.2 Astra, #3 Gemini 4 Carbon, #6 GPT-6.2 Sol.

Interactive version: theaggregate.ai/benchmark?slug=policybench-us-exact-match-tanf · How It Works · Data refreshed daily, snapshot 2026-11-17.