Benchmark Almanac

Vals Index v2.2

Claude Sonnet 5.5 70.8% leads since Oct 2026 In use

Agents are scored on eight finance, coding, legal, and tax benchmarks, combined with weights from each sector's share of US GDP.

Version 2.2 from 19 October 2026: (8.0 × finance + 5.6 × coding + 1.2 × legal + 0.5 × tax) / 15.3. Finance averages Finance Agent v2.1 and EMB v1.1. Coding averages Terminal-Bench 4, Vibe Code Bench v1.1 and Code Migration v1.1. Legal averages Legal Research and HLAB-Vals-1.1. Tax is Tax Agent Bench (all-pass accuracy). Six of the eight components are private. Earlier versions have their own records.

Made by Vals AI Released Oct 2026 Composite work indices Public tasks plus a private test set Last in a launch post Oct 2026 Last model added Oct 2026 Scores refreshed 2 Nov 2026
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Leaderboard vals.ai
32
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab

Claude Sonnet 5.5 leads with 70.8%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

32 models. Scroll sideways for the rest.

  1. 70.8 Claude Sonnet 5.5
  2. 70.5 Gemini 4 Carbon
  3. 70.2 Claude Opus 5.5
  4. 69.2 Gemini 4 Argon
  5. 66.8 Muse Wonder Full
  6. 65.0 GPT-6 Astra
  7. 64.9 Muse Wonder
  8. 63.9 GPT-6.1 Sol
  9. 62.5 Kimi K3.1 Pro
  10. 61.3 DeepSeek V4.1 Flash 1017
  11. 60.5 GPT-6.1 Terra
  12. 59.7 GLM-5.3
  13. 59.0 Qwen3.8 Max (0902)
  14. 58.7 Claude Haiku 5.5
  15. 58.6 Grok 4.7
  16. 57.8 MiMo-V2.6 Pro
  17. 56.5 Mistral Large 4 (Le Chonk, open weights)
  18. 56.0 Gemini 3.8 Flash
  19. 54.8 MiMo-V2.6 Flash
  20. 54.8 Step 5 Preview
  21. 53.8 MiniMax-M3.1-Flash
  22. 53.2 Tencent Hy4 preview
  23. 53.0 DeepSeek V4 Pro 0813
  24. 53.0 Kimi K3
  25. 52.3 GPT-6 Luna
  26. 50.9 GLM-5.3-Flash
  27. 48.6 Muse Spark Open
  28. 43.7 Mercury 3
  29. 43.4 Beam 1.0
  30. 29.8 Nemotron 3.5 Super VL
  31. 29.0 Inkling
  32. 27.6 Inkling-Small
  1. 1 Claude Sonnet 5.5 70.8%
  2. 2 Gemini 4 Carbon 70.5%
  3. 3 Claude Opus 5.5 70.2%
  4. 4 Gemini 4 Argon 69.2%
  5. 5 Muse Wonder Full 66.8%
  6. 6 GPT-6 Astra 65%
  7. 7 Muse Wonder 64.9%
  8. 8 GPT-6.1 Sol 63.9%
  9. 9 Kimi K3.1 Pro 62.5%
  10. 10 DeepSeek V4.1 Flash 1017 61.3%
  11. 11 GPT-6.1 Terra 60.5%
  12. 12 GLM-5.3 59.7%
  13. 13 Qwen3.8 Max (0902) 59%
  14. 14 Claude Haiku 5.5 58.7%
  15. 15 Grok 4.7 58.6%
  16. 16 MiMo-V2.6 Pro 57.8%
  17. 17 Mistral Large 4 (Le Chonk, open weights) 56.5%
  18. 18 Gemini 3.8 Flash 56%
  19. 19 MiMo-V2.6 Flash 54.8%
  20. 20 Step 5 Preview 54.8%
  21. 21 MiniMax-M3.1-Flash 53.8%
  22. 22 Tencent Hy4 preview 53.2%
  23. 23 DeepSeek V4 Pro 0813 53%
  24. 24 Kimi K3 53%
  25. 25 GPT-6 Luna 52.3%
  26. 26 GLM-5.3-Flash 50.9%
  27. 27 Muse Spark Open 48.6%
  28. 28 Mercury 3 43.7%
  29. 29 Beam 1.0 43.4%
  30. 30 Nemotron 3.5 Super VL 29.8%
  31. 31 Inkling 29%
  32. 32 Inkling-Small 27.6%
#ModelScoreSettingDateSource
1 Claude Sonnet 5.5 70.8% Sep 2026 Official leaderboard
2 Gemini 4 Carbon 70.5% Oct 2026 Official leaderboard
3 Claude Opus 5.5 70.2% Sep 2026 Official leaderboard
4 Gemini 4 Argon 69.2% Sep 2026 Official leaderboard
5 Muse Wonder Full 66.8% Oct 2026 Official leaderboard
6 GPT-6 Astra 65% Sep 2026 Official leaderboard
7 Muse Wonder 64.9% Oct 2026 Official leaderboard
8 GPT-6.1 Sol 63.9% Sep 2026 Official leaderboard
9 Kimi K3.1 Pro 62.5% Oct 2026 Official leaderboard
10 DeepSeek V4.1 Flash 1017 61.3% Oct 2026 Official leaderboard
11 GPT-6.1 Terra 60.5% Oct 2026 Official leaderboard
12 GLM-5.3 59.7% Aug 2026 Official leaderboard
13 Qwen3.8 Max (0902) 59% Sep 2026 Official leaderboard
14 Claude Haiku 5.5 58.7% Oct 2026 Official leaderboard
15 Grok 4.7 58.6% Sep 2026 Official leaderboard
16 MiMo-V2.6 Pro 57.8% Sep 2026 Official leaderboard
17 Mistral Large 4 (Le Chonk, open weights) 56.5% Oct 2026 Official leaderboard
18 Gemini 3.8 Flash 56% Sep 2026 Official leaderboard
19 MiMo-V2.6 Flash 54.8% Sep 2026 Official leaderboard
20 Step 5 Preview 54.8% Sep 2026 Official leaderboard
21 MiniMax-M3.1-Flash 53.8% Oct 2026 Official leaderboard
22 Tencent Hy4 preview 53.2% Aug 2026 Official leaderboard
23 DeepSeek V4 Pro 0813 53% Aug 2026 Official leaderboard
24 Kimi K3 53% Jul 2026 Official leaderboard
25 GPT-6 Luna 52.3% Sep 2026 Official leaderboard
26 GLM-5.3-Flash 50.9% Aug 2026 Official leaderboard
27 Muse Spark Open 48.6% Oct 2026 Official leaderboard
28 Mercury 3 43.7% Oct 2026 Official leaderboard
29 Beam 1.0 43.4% Oct 2026 Official leaderboard
30 Nemotron 3.5 Super VL 29.8% Oct 2026 Official leaderboard
31 Inkling 29% Jul 2026 Official leaderboard
32 Inkling-Small 27.6% Jul 2026 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

The best score, 70.8%, costs $19 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Wonder Full comes within 4.1 points of the best for $6.81 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Xiaomi Corp Anthropic OpenAI Meta DeepSeek Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Sonnet 5.5 Muse Wonder Full DeepSeek V4.1 Flash 1017 MiMo-V2.6 Flash 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Sonnet 5.5 GPT-6.1 Sol
ModelEffortHarnessScoreCost per taskOutput tokens
Claude Sonnet 5.5 70.8% $19
Gemini 4 Carbon 70.5% $11
Claude Opus 5.5 70.2% $28
Gemini 4 Argon 69.2% $13
Muse Wonder Full 66.8% $6.81
GPT-6 Astra 65% $21
Muse Wonder 64.9% $11
GPT-6.1 Sol 63.9% $3.61
Kimi K3.1 Pro 62.5% $6.44
DeepSeek V4.1 Flash 1017 61.3% $0.66
GPT-6.1 Terra 60.5% $2.94
GLM-5.3 59.7% $8.52
Qwen3.8 Max (0902) 59% $8.73
Claude Haiku 5.5 58.7% $3.17
Grok 4.7 58.6% $12
MiMo-V2.6 Pro 57.8% $0.43
Mistral Large 4 (Le Chonk, open weights) 56.5% $3.72
Gemini 3.8 Flash 56% $5.03
MiMo-V2.6 Flash 54.8% $0.22
Step 5 Preview 54.8% $2.73
MiniMax-M3.1-Flash 53.8% $0.66
Tencent Hy4 preview 53.2% $1.57
DeepSeek V4 Pro 0813 53% $2.50
Kimi K3 53% $7.17
GPT-6 Luna 52.3% $0.45
GLM-5.3-Flash 50.9% $1.13
Muse Spark Open 48.6% $4.72
Mercury 3 43.7% $0.48
Beam 1.0 43.4%
Nemotron 3.5 Super VL 29.8% $0.63
Inkling 29% $1.85
Inkling-Small 27.6% $0.40

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard 1 leader
Claude Sonnet 5.5 70.8%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Sonnet 5.5 GPT-6 Astra GLM-5.3 DeepSeek V4 Pro 0813 Inkling 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Sonnet 5.5 GLM-5.3

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Google DeepMind
DateLaunch postScoreAs the post writes it
28 Oct 2026 Gemini 4 Carbon 70.4 Vals Index v2.2

Versions

Replaces Vals Index v2.1

More in Economic work

All 91 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Qwen3.8-Omni-Flash 75.3%
  • OfficeQA In use Grounded reasoning over a large corpus of U.S. Treasury Bulletin documents: locate tables and compute precise numeric answers. Claude Opus 4.7 86.3%