Share promptsSign inCreate team library
← All benchmarks

Overall work AI benchmarks

One number for agentic work across finance, coding, law and tax.

Vals Index

One score for agentic work across finance, coding, legal and tax tasks, weighted by each sector’s share of US GDP.

Run by Vals AI. Leaderboard as of 19 Oct 2026 · retrieved 21 Oct 2026.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • DeepSeek
  • Z.ai
  • Moonshot
  • MiniMax
  • Xiaomi
  • Other
Vals Index: Accuracy vs Cost per test (USD)Scatter plot of 22 models. Leader on Accuracy: Claude Sonnet 5.5 at 70.85%. Full values in the table below.$.1$1$10$10020%40%60%80%Cost per test (USD) (log scale) →↑ AccuracyInkling Small (Other): Cost per test (USD) $0.40, Accuracy 27.59%·Inkling (Other): Cost per test (USD) $1.85, Accuracy 28.96%·GLM 5.3 Flash (Z.ai): Cost per test (USD) $1.13, Accuracy 50.86%ZGPT-6 Luna (OpenAI): Cost per test (USD) $0.45, Accuracy 52.32%OKimi K3 (Moonshot): Cost per test (USD) $7.17, Accuracy 52.95%KDeepSeek V4 Pro 0813 (DeepSeek): Cost per test (USD) $2.50, Accuracy 52.97%DHy4 Preview (Other): Cost per test (USD) $1.45, Accuracy 53.21%·MiniMax-M3.1-Flash (MiniMax): Cost per test (USD) $0.66, Accuracy 53.78%MMMiMo V2.6 Flash (Xiaomi): Cost per test (USD) $0.22, Accuracy 54.83%XiGemini 3.8 Flash (Google): Cost per test (USD) $6.06, Accuracy 55.95%GMiMo V2.6 Pro (Xiaomi): Cost per test (USD) $0.43, Accuracy 57.81%XiGrok 4.7 (xAI): Cost per test (USD) $12, Accuracy 58.56%XClaude Haiku 5.5 (Anthropic): Cost per test (USD) $3.17, Accuracy 58.71%AGLM 5.3 (Z.ai): Cost per test (USD) $8.52, Accuracy 59.73%ZDeepSeek V4.1 Flash 1017 (DeepSeek): Cost per test (USD) $0.66, Accuracy 61.26%DGPT-6.1 Sol (OpenAI): Cost per test (USD) $3.61, Accuracy 63.94%OWonder (Meta): Cost per test (USD) $11, Accuracy 64.88%MGPT-6 Astra (OpenAI): Cost per test (USD) $21, Accuracy 65.02%OWonder Full (Meta): Cost per test (USD) $6.81, Accuracy 66.79%MGemini 4 Argon (Google): Cost per test (USD) $13, Accuracy 69.24%GClaude Opus 5.5 (Anthropic): Cost per test (USD) $28, Accuracy 70.18%AClaude Sonnet 5.5 (Anthropic): Cost per test (USD) $19, Accuracy 70.85%AClaude Sonnet 5.5Claude Opus 5.5Gemini 4 Argon
Vals Index: top 15 by AccuracyRanked bars. Leader: Claude Sonnet 5.5 at 70.85%. Full values in the table below.0%20%40%60%80%100%Claude Sonnet 5.5 (Anthropic): 70.85%AClaude Sonnet 5.570.85%Claude Opus 5.5 (Anthropic): 70.18%AClaude Opus 5.570.18%Gemini 4 Argon (Google): 69.24%GGemini 4 Argon69.24%Wonder Full (Meta): 66.79%MWonder Full66.79%GPT-6 Astra (OpenAI): 65.02%OGPT-6 Astra65.02%Wonder (Meta): 64.88%MWonder64.88%GPT-6.1 Sol (OpenAI): 63.94%OGPT-6.1 Sol63.94%DeepSeek V4.1 Flash 1017 (DeepSeek): 61.26%DDeepSeek V4.1 Flash 101761.26%GLM 5.3 (Z.ai): 59.73%ZGLM 5.359.73%Claude Haiku 5.5 (Anthropic): 58.71%AClaude Haiku 5.558.71%Grok 4.7 (xAI): 58.56%XGrok 4.758.56%MiMo V2.6 Pro (Xiaomi): 57.81%XiMiMo V2.6 Pro57.81%Gemini 3.8 Flash (Google): 55.95%GGemini 3.8 Flash55.95%MiMo V2.6 Flash (Xiaomi): 54.83%XiMiMo V2.6 Flash54.83%MiniMax-M3.1-Flash (MiniMax): 53.78%MMMiniMax-M3.1-Flash53.78%
Accuracy
GDP-weighted accuracy across the Vals agentic benchmarks.
Cost per test (USD)
Average API cost to run the index tasks, as published.
Vals Index scores
RankModelProviderAccuracyCost per test (USD)Price in / out ($/1M tokens)
1 Claude Sonnet 5.5Anthropic70.85%$19$2 / $10
2 Claude Opus 5.5Anthropic70.18%$28$4 / $20
3 Gemini 4 ArgonGoogle69.24%$13$4 / $20
4 Wonder FullMeta66.79%$6.81$4 / $20
5 GPT-6 AstraOpenAI65.02%$21$10 / $50
6 WonderMeta64.88%$11$4 / $20
7 GPT-6.1 SolOpenAI63.94%$3.61$2 / $10
8 DeepSeek V4.1 Flash 1017DeepSeek61.26%$0.66$0.3 / $1.2
9 GLM 5.3Z.ai59.73%$8.52$1.4 / $4.4
10 Claude Haiku 5.5Anthropic58.71%$3.17$0.1 / $0.5
Show 13 more rows
RankModelProviderAccuracyCost per test (USD)Price in / out ($/1M tokens)
11 Grok 4.7xAI58.56%$12$2 / $6
12 MiMo V2.6 ProXiaomi57.81%$0.43$0.435 / $0.87
13 Gemini 3.8 FlashGoogle55.95%$6.06$1.5 / $7.5
14 MiMo V2.6 FlashXiaomi54.83%$0.22$0.14 / $0.28
15 MiniMax-M3.1-FlashMiniMax53.78%$0.66$0.15 / $0.6
16 Hy4 PreviewOther53.21%$1.45$0.834 / $2.501
17 DeepSeek V4 Pro 0813DeepSeek52.97%$2.50$1.32 / $3.96
18 Kimi K3Moonshot52.95%$7.17$3 / $15
19 GPT-6 LunaOpenAI52.32%$0.45$0.1 / $0.5
20 GLM 5.3 FlashZ.ai50.86%$1.13$0.075 / $0.25
21 Beam 1.0Other43.42%–N/A
22 InklingOther28.96%$1.85$1 / $4.05
23 Inkling SmallOther27.59%$0.40$0.3 / $1.2

Source: Vals AI leaderboard ↗

Top model here: Claude Sonnet 5.5 (Anthropic). Try prompts for Claude →

ChatJudge does not run these tests. Scores are copied as published by each source, with a link and the date we read them, and only change when we refresh from the source. Different leaderboards use different settings, so compare models within one chart, not across charts. Spotted a mistake? Tell us.