Benchmarks · Vals AI
Public Benefits Bench
Measures whether AI can help people navigate US SNAP benefits: questions on benefits workflows.
- Models measured
- 50
- Top
- 76.93 %Claude Opus 5within the standard error of ±1.10
- Version
- 1.1
- Measured
- 12 October 2026
01
Ranking
At the top: Top 15
Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.
- 1Claude Opus 576.93%
- 2Claude Fable 5.174.90%
- 3Wonder Full73.75%
- 4Claude Opus 5.570.64%
- 5Claude Fable 570.43%
- 6Wonder70.23%
- 7Gemini 4 Argon69.76%
- 8MiMo V2.6 Pro68.94%
- 9Hy4 Preview68.61%
- 10GLM 5.368.54%
- 11Muse Spark 1.268.47%
- 12Kimi K368.20%
- 13Claude Opus 4.868.13%
- 14MiMo V2.6 Flash67.59%
- 15Claude Sonnet 5.567.19%
Source: Vals AI (Public Benefits Bench) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.
| # Model | Score | Cost/task | Gate | PrudentScore |
|---|---|---|---|---|
| 1Claude Opus 5Anthropic | 76.93%± 1.10 | $3.95 | Passes | 72 |
| 2Claude Fable 5.1Anthropic | 74.90%± 1.13 | $7.48 | Rejected | no score |
| 3Wonder FullMeta | 73.75%± 1.15 | $4.64 | Rejected | no score |
| 4Claude Opus 5.5Anthropic | 70.64%± 1.19 | $5.29 | Passes | 63 |
| 5Claude Fable 5Anthropic | 70.43%± 1.19 | $4.59 | Rejected | no score |
| 6WonderMeta | 70.23%± 1.19 | $2.46 | Rejected | no score |
| 7Gemini 4 ArgonGoogle | 69.76%± 0.00 | $2.79 | Passes | 79 |
| 8MiMo V2.6 ProXiaomi | 68.94%± 1.20 | $0.08 | Rejected | no score |
| 9Hy4 PreviewTencent | 68.61%± 1.21 | $0.25 | Rejected | no score |
| 10GLM 5.3Z.ai | 68.54%± 1.21 | $0.95 | Rejected | no score |
| 11Muse Spark 1.2Meta | 68.47%± 1.21 | $0.59 | Rejected | no score |
| 12Kimi K3Moonshot AI | 68.20%± 1.21 | $0.91 | Rejected | no score |
| 13Claude Opus 4.8Anthropic | 68.13%± 1.21 | $1.89 | Passes | 62 |
| 14MiMo V2.6 FlashXiaomi | 67.59%± 1.22 | $0.03 | Rejected | no score |
| 15Claude Sonnet 5.5Anthropic | 67.19%± 1.22 | $3.76 | Passes | 62 |
| 16Qwen 3.8 MaxAlibaba | 67.12%± 1.22 | $0.99 | Not yet assessed | no score |
| 17Grok 4.6xAI (SpaceXAI) | 66.85%± 1.23 | $0.92 | Rejected | no score |
| 18GPT-5.6 SolOpenAI | 66.51%± 1.23 | $8.85 | Passes | 57 |
| 19Ember-1Fireworks AI | 66.17%± 1.23 | $1.02 | Not yet assessed | no score |
| 20Claude Sonnet 5Anthropic | 66.03%± 1.23 | $1.29 | Passes | 59 |
| 21Grok 4.7xAI (SpaceXAI) | 65.63%± 1.24 | $2.25 | Rejected | no score |
| 22Gemini 3.8 FlashGoogle | 65.29%± 1.24 | $1.01 | Passes | 62 |
| 23DeepSeek V4.1 FlashDeepSeek | 64.28%± 1.25 | $0.07 | Rejected | no score |
| 24MiniMax-M3MiniMax | 64.14%± 1.25 | $0.25 | Rejected | no score |
| 25Claude Haiku 5.5Anthropic | 63.19%± 1.25 | $0.53 | Passes | 59 |
| 26DeepSeek V4 Pro (2026-04-23)DeepSeek | 62.92%± 1.26 | $0.46 | Rejected | no score |
| 27Claude Sonnet 4.6Anthropic | 62.45%± 1.26 | $1.11 | Passes | 55 |
| 28GPT-5.6 TerraOpenAI | 62.38%± 1.26 | $1.20 | Passes | 53 |
| 29GLM 5.1Z.ai | 61.84%± 1.26 | $0.34 | Rejected | no score |
| 30Muse Spark 1.1Meta | 61.30%± 1.27 | $0.28 | Rejected | no score |
| 31GPT-5.6 LunaOpenAI | 61.16%± 1.27 | $0.27 | Passes | 56 |
| 32GPT 5.5OpenAI | 60.89%± 1.27 | $3.97 | Passes | 51 |
| 33Gemini 3.5 FlashGoogle | 59.47%± 1.28 | $1.11 | Passes | 47 |
| 34Inkling SmallThinking Machines | 59.41%± 1.28 | $0.10 | Rejected | no score |
| 35GPT-6.1 SolOpenAI | 59.34%± 1.28 | $0.69 | Passes | 59 |
| 36InklingThinking Machines | 58.46%± 1.28 | $0.37 | Rejected | no score |
| 37Step 5 PreviewStepFun | 58.12%± 1.28 | $0.50 | Rejected | no score |
| 38GPT-6 LunaOpenAI | 57.65%± 1.28 | $0.22 | Passes | 54 |
| 39=GPT-6 SolOpenAI | 56.63%± 1.29 | $2.49 | Passes | 45 |
| 39=Kimi K2.6Moonshot AI | 56.63%± 1.29 | $0.91 | Rejected | no score |
| 41Gemini 3.6 FlashGoogle | 56.56%± 1.29 | $0.39 | Passes | 46 |
| 42Claude Haiku 4.5 (Nonthinking)Anthropic | 54.26%± 1.30 | $0.22 | Passes | no score |
| 43Gemini 3.1 Pro Preview (02/26)Google | 53.79%± 1.30 | $0.38 | Rejected | no score |
| 44Gemini 3.5 Flash LiteGoogle | 51.83%± 1.30 | $0.10 | Passes | 38 |
| 45Grok 4.3xAI (SpaceXAI) | 51.69%± 1.30 | $0.26 | Rejected | no score |
| 46Ling 3.0 FlashAnt Group | 50.74%± 1.30 | $0.05 | Rejected | no score |
| 47Mercury 2.5Inception Labs | 45.26%± 1.29 | $0.07 | Not yet assessed | no score |
| 48Grok 4.1 Fast (Reasoning)xAI (SpaceXAI) | 44.79%± 1.29 | $0.03 | Rejected | no score |
| 49Laguna M.1Poolside | 43.84%± 1.29 | $0.20 | Rejected | no score |
| 50Laguna XS.2Poolside | 41.41%± 1.28 | $0.11 | Rejected | no score |
Source: Vals AI (Public Benefits Bench) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.
Vals AI · Version 1.1 · Measured 12 October 2026 · Go to the source. The PrudentScore is a composition by Prudai B.V. based on publicly published figures and does not represent the view or endorsement of Vals AI, Harvey or Artificial Analysis. Figures from Artificial Analysis appear in running text only, with attribution and a link, as that party's terms require [9].
Other benchmarks
- Legal Research Bench
- Harvey LAB (Vals implementation)
- LegalBench (Vals leaderboard)
- CaseLaw v2
- TaxEval v2
- Tax Agent Bench
- CorpFin v2
- Immigration Bench
- Access to Justice
- Vals Legal Index
- Vals Index
- GPQA Diamond (Vals-run)
- MMLU Pro (Vals-run)
- Artificial Analysis Intelligence Index
Claude and Anthropic, GPT and OpenAI, Gemini and Google, Grok and xAI, Kimi and Moonshot AI, Muse and Meta, Mistral, DeepSeek, GLM and Z.ai, Harvey, Vals AI and Artificial Analysis are trademarks of their respective owners. They are named for identification only and this implies no partnership, sponsorship or endorsement.