Public Benefits Bench (Vals AI) leaderboard
- Claude Opus 5 leads
- 76.9%
- models measured
- 46
- lead over #2
- +2.0 pts
- median
- 63.7%
Leader on Public Benefits Bench (Vals AI) · read October 22
On October 22, 2026, 46 of our catalog's models had a Public Benefits Bench (Vals AI) result, and Anthropic Claude Opus 5 led with 76.9%, ahead of Anthropic Claude Fable 5.1 at 74.9%.
Guidance on SNAP benefits across the steps of a benefits workflow. Share correct, as Vals AI grades it.
Role in the index
Public Benefits Bench (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 76.9% | $3.95 | 83.0 | $25.00 | |
| 2 | Claude Fable 5.1Anthropic | 74.9% | $7.48 | 84.2 | $50.00 | |
| 3 | Claude Opus 5.5Anthropic | 70.6% | $5.95 | 88.8 | $20.00 | |
| 4 | Claude Fable 5Anthropic | 70.4% | $4.59 | 79.3 | $50.00 | |
| 5 | Muse WonderMeta | 70.2% | $2.46 | 71.9 | $20.00 | |
| 6 | Gemini 4 ArgonGoogle | 69.8% | $2.79 | 81.0 | $10.00 | |
| 7 | MiMo-V2.6-ProXiaomi | 68.9% | $0.08 | 65.4 | $0.87 | |
| 8 | Hy4 previewTencent | 68.6% | $0.25 | 44.3 | $2.501 | |
| 9 | GLM 5.3Z.ai | 68.5% | $0.95 | 70.6 | $4.40 | |
| 10 | Muse Spark 1.2Meta | 68.5% | $0.59 | 55.7 | $4.25 | |
| 11 | Kimi K3Moonshot AI | 68.2% | $0.91 | 62.3 | $15.00 | |
| 12 | Claude Opus 4.8Anthropic | 68.1% | $1.89 | 65.0 | $25.00 | |
| 13 | MiMo-V2.6-FlashXiaomi | 67.6% | $0.03 | 54.7 | $0.28 | |
| 14 | Claude Sonnet 5.5Anthropic | 67.2% | $4.25 | 87.3 | $10.00 | |
| 15 | Qwen3.8 Max (0902)Qwen | 67.1% | $0.99 | 62.4 | $6.00 | |
| 16 | Grok 4.6xAI | 66.8% | $0.92 | 65.7 | $6.00 | |
| 17 | GPT-5.6 SolOpenAI | 66.5% | $8.85 | 73.5 | $20.00 | |
| 18 | Ember-1Fireworks | 66.2% | $1.02 | 52.6 | $15.00 | |
| 19 | Claude Sonnet 5Anthropic | 66.0% | $1.29 | 57.7 | $10.00 | |
| 20 | Grok 4.7xAI | 65.6% | $2.25 | 66.7 | $6.00 | |
| 21 | Gemini 3.8 FlashGoogle | 65.3% | $1.01 | 61.6 | $3.75 | |
| 22 | DeepSeek V4.1 FlashDeepSeek | 64.3% | $0.07 | 62.5 | $0.84 | |
| 23 | MiniMax M3MiniMax | 64.1% | $0.25 | 39.2 | $1.20 | |
| 24 | Claude Haiku 5.5Anthropic | 63.2% | $0.53 | 70.7 | $0.50 | |
| 25 | DeepSeek V4 Pro 0423DeepSeek | 62.9% | $0.46 | 38.7 | $2.976 | |
| 26 | Claude Sonnet 4.6Anthropic | 62.5% | $1.11 | 57.3 | $15.00 | |
| 27 | GPT-5.6 TerraOpenAI | 62.4% | $1.20 | 63.9 | $12.00 | |
| 28 | GLM 5.1Z.ai | 61.8% | $0.34 | 49.2 | $4.40 | |
| 29 | Muse Spark 1.1Meta | 61.3% | $0.28 | 70.3 | $4.25 | |
| 30 | GPT-5.6 LunaOpenAI | 61.2% | $0.27 | 58.0 | $1.20 | |
| 31 | GPT-5.5OpenAI | 60.9% | $3.97 | 71.3 | $30.00 | |
| 32 | Gemini 3.5 FlashGoogle | 59.5% | $1.11 | 39.8 | $9.00 | |
| 33 | T Inkling SmallThinking Machines | 59.4% | $0.10 | 27.3 | $1.20 | |
| 34 | GPT-6.1 SolOpenAI | 59.3% | $0.69 | 82.4 | $10.00 | |
| 35 | T InklingThinking Machines | 58.5% | $0.37 | 31.5 | $4.05 | |
| 36 | Step 5 PreviewStepFun | 58.1% | $0.50 | 61.1 | $2.70 | |
| 37 | GPT-6 LunaOpenAI | 57.6% | $0.22 | 57.3 | $0.50 | |
| 38 | Kimi K2.6Moonshot AI | 56.6% | $0.91 | 52.1 | $4.00 | |
| 39 | GPT-6 SolOpenAI | 56.6% | $2.49 | 76.0 | $10.00 | |
| 40 | Gemini 3.6 FlashGoogle | 56.6% | $0.39 | 62.5 | $3.75 | |
| 41 | Claude Haiku 4.5Anthropic | 54.3% | $0.22 | 35.6 | $5.00 | |
| 42 | Gemini 3.1 Pro PreviewGoogle | 53.8% | $0.38 | 32.3 | $12.00 | |
| 43 | Gemini 3.5 Flash LiteGoogle | 51.8% | $0.10 | 49.0 | $2.50 | |
| 44 | Grok 4.3xAI | 51.7% | $0.26 | 37.4 | $2.50 | |
| 45 | Ling 3.0 FlashinclusionAI | 50.7% | $0.05 | 29.6 | $0.1215 | |
| 46 | Mercury 2.5Inception | 45.3% | $0.07 | 1.6 | $0.15 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars.