Public Benefits Bench v1

Can AI help people navigate SNAP benefits?

ModelsModels with a result.12
Top resultBest result on this test.71.7%Claude Fable 5
Top-3 spreadPoints from first to third.11.0 pts
UpdatedDate the source changed the results.9 Jun 2026

Result and cost

020406080100$0.03$0.1$0.3$1$3RESULTUS DOLLARS PER TASK · LOG SCALE

Results

12 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01Claude Fable 571.7%$1.0517 min
02Claude Opus 4.862.1%$1.3210 min
03MiniMax M360.7%$0.19912 min
04Claude Sonnet 4.658.5%$0.81613 min
05Gemini 3.5 Flash58.0%$0.7765 min
06GLM-5.157.9%$0.3397 min
07DeepSeek V4 Pro 042357.6%$0.3616 min
08GPT-5.557.2%$2.0226 min
09Gemini 3.1 Pro Preview50.9%$0.3013 min
10Grok 4.350.1%$1.217 min
11Claude Haiku 4.549.5%$0.1953 min
12Grok 4.1 Fast (Reasoning)44.3%$0.0653 min