Benchmark Almanac

Vals Finance Index

Gemini 4 Carbon 67% leads since Oct 2026 Not in model cards

Agents are scored on Finance Agent v2.1, Excel Modeling Benchmark v1.1 and Tax Agent Bench, combined with weights from each sub-sector share of US GDP.

Securities work (1.6% of GDP) averages Finance Agent and EMB. Tax work (0.5%) uses Tax Agent Bench all-pass accuracy. 35 models on 2026-10-29.

Made by Vals AI Released Oct 2026 Composite work indices Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
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
35
Models scored on its own leaderboard

Gemini 4 Carbon leads with 67%

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.

35 models. Scroll sideways for the rest.

  1. 67.0 Gemini 4 Carbon
  2. 65.6 Claude Opus 5.5
  3. 64.9 Gemini 4 Argon
  4. 64.5 Muse Wonder Full
  5. 64.5 Claude Sonnet 5.5
  6. 61.7 Qwen3.8 Max (0902)
  7. 61.2 Muse Wonder
  8. 60.9 DeepSeek V4.1 Flash 1017
  9. 60.5 GLM-5.3
  10. 60.5 Gemini 3.8 Flash
  11. 58.5 Grok 4.7
  12. 58.0 Kimi K3
  13. 57.9 MiMo-V2.6 Pro
  14. 57.5 Tencent Hy4 preview
  15. 57.5 Claude Haiku 5.5
  16. 57.1 MiMo-V2.6 Flash
  17. 57.1 DeepSeek V4.1 Flash
  18. 57.0 Mistral Large 4 (Le Chonk, open weights)
  19. 56.9 Mistral Large 4
  20. 56.0 Kimi K3.1 Pro
  21. 55.9 MiniMax-M3.1-Flash
  22. 55.8 GLM-5.3-Flash
  23. 55.8 GPT-6.1 Sol
  24. 55.6 GPT-6 Astra
  25. 55.4 Step 5 Preview
  26. 53.3 GPT-6.1 Terra
  27. 53.0 GPT-6 Luna
  28. 53.0 Muse Spark Open
  29. 52.9 DeepSeek V4 Pro 0813
  30. 48.4 MiniMax-M3
  31. 47.2 Mercury 3
  32. 46.5 Beam 1.0
  33. 41.0 Nemotron 3.5 Super VL
  34. 38.3 Inkling
  35. 36.3 Inkling-Small
  1. 1 Gemini 4 Carbon 67%
  2. 2 Claude Opus 5.5 65.6%
  3. 3 Gemini 4 Argon 64.9%
  4. 4 Muse Wonder Full 64.5%
  5. 5 Claude Sonnet 5.5 64.5%
  6. 6 Qwen3.8 Max (0902) 61.7%
  7. 7 Muse Wonder 61.2%
  8. 8 DeepSeek V4.1 Flash 1017 60.9%
  9. 9 GLM-5.3 60.5%
  10. 10 Gemini 3.8 Flash 60.5%
  11. 11 Grok 4.7 58.5%
  12. 12 Kimi K3 58%
  13. 13 MiMo-V2.6 Pro 57.9%
  14. 14 Tencent Hy4 preview 57.5%
  15. 15 Claude Haiku 5.5 57.5%
  16. 16 MiMo-V2.6 Flash 57.1%
  17. 17 DeepSeek V4.1 Flash 57.1%
  18. 18 Mistral Large 4 (Le Chonk, open weights) 57%
  19. 19 Mistral Large 4 56.9%
  20. 20 Kimi K3.1 Pro 56%
  21. 21 MiniMax-M3.1-Flash 55.9%
  22. 22 GLM-5.3-Flash 55.8%
  23. 23 GPT-6.1 Sol 55.8%
  24. 24 GPT-6 Astra 55.6%
  25. 25 Step 5 Preview 55.4%
  26. 26 GPT-6.1 Terra 53.3%
  27. 27 GPT-6 Luna 53%
  28. 28 Muse Spark Open 53%
  29. 29 DeepSeek V4 Pro 0813 52.9%
  30. 30 MiniMax-M3 48.4%
  31. 31 Mercury 3 47.2%
  32. 32 Beam 1.0 46.5%
  33. 33 Nemotron 3.5 Super VL 41%
  34. 34 Inkling 38.3%
  35. 35 Inkling-Small 36.3%
#ModelScoreSettingDateSource
1 Gemini 4 Carbon 67% Oct 2026 Official leaderboard
2 Claude Opus 5.5 65.6% Sep 2026 Official leaderboard
3 Gemini 4 Argon 64.9% Sep 2026 Official leaderboard
4 Muse Wonder Full 64.5% Oct 2026 Official leaderboard
5 Claude Sonnet 5.5 64.5% Sep 2026 Official leaderboard
6 Qwen3.8 Max (0902) 61.7% Sep 2026 Official leaderboard
7 Muse Wonder 61.2% Oct 2026 Official leaderboard
8 DeepSeek V4.1 Flash 1017 60.9% Oct 2026 Official leaderboard
9 GLM-5.3 60.5% Aug 2026 Official leaderboard
10 Gemini 3.8 Flash 60.5% Sep 2026 Official leaderboard
11 Grok 4.7 58.5% Sep 2026 Official leaderboard
12 Kimi K3 58% Jul 2026 Official leaderboard
13 MiMo-V2.6 Pro 57.9% Sep 2026 Official leaderboard
14 Tencent Hy4 preview 57.5% Aug 2026 Official leaderboard
15 Claude Haiku 5.5 57.5% Oct 2026 Official leaderboard
16 MiMo-V2.6 Flash 57.1% Sep 2026 Official leaderboard
17 DeepSeek V4.1 Flash 57.1% Sep 2026 Official leaderboard
18 Mistral Large 4 (Le Chonk, open weights) 57% Oct 2026 Official leaderboard
19 Mistral Large 4 56.9% Oct 2026 Official leaderboard
20 Kimi K3.1 Pro 56% Oct 2026 Official leaderboard
21 MiniMax-M3.1-Flash 55.9% Oct 2026 Official leaderboard
22 GLM-5.3-Flash 55.8% Aug 2026 Official leaderboard
23 GPT-6.1 Sol 55.8% Sep 2026 Official leaderboard
24 GPT-6 Astra 55.6% Sep 2026 Official leaderboard
25 Step 5 Preview 55.4% Sep 2026 Official leaderboard
26 GPT-6.1 Terra 53.3% Oct 2026 Official leaderboard
27 GPT-6 Luna 53% Sep 2026 Official leaderboard
28 Muse Spark Open 53% Oct 2026 Official leaderboard
29 DeepSeek V4 Pro 0813 52.9% Aug 2026 Official leaderboard
30 MiniMax-M3 48.4% Jun 2026 Official leaderboard
31 Mercury 3 47.2% Oct 2026 Official leaderboard
32 Beam 1.0 46.5% Oct 2026 Official leaderboard
33 Nemotron 3.5 Super VL 41% Oct 2026 Official leaderboard
34 Inkling 38.3% Jul 2026 Official leaderboard
35 Inkling-Small 36.3% Jul 2026 Official leaderboard

The best score, 67%, costs $6.46 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Wonder Full comes within 2.5 points of the best for $3.14 a task, about half the cost. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Xiaomi Corp Meta DeepSeek Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Carbon DeepSeek V4.1 Flash 1017 MiMo-V2.6 Flash 0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Carbon DeepSeek V4.1 Flash 1017
ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Carbon 67% $6.46
Claude Opus 5.5 65.6% $12
Gemini 4 Argon 64.9% $6.53
Muse Wonder Full 64.5% $3.14
Claude Sonnet 5.5 64.5% $8.33
Qwen3.8 Max (0902) 61.7% $3.83
Muse Wonder 61.2% $3.37
DeepSeek V4.1 Flash 1017 60.9% $0.42
GLM-5.3 60.5% $2.21
Gemini 3.8 Flash 60.5% $3.70
Grok 4.7 58.5% $3.51
Kimi K3 58% $2.58
MiMo-V2.6 Pro 57.9% $0.22
Tencent Hy4 preview 57.5% $0.68
Claude Haiku 5.5 57.5% $1.39
MiMo-V2.6 Flash 57.1% $0.07
DeepSeek V4.1 Flash 57.1% $0.25
Mistral Large 4 (Le Chonk, open weights) 57% $0.56
Mistral Large 4 56.9% $1.46
Kimi K3.1 Pro 56% $2.51
MiniMax-M3.1-Flash 55.9% $0.25
GLM-5.3-Flash 55.8% $0.31
GPT-6.1 Sol 55.8% $1.36
GPT-6 Astra 55.6% $6.48
Step 5 Preview 55.4% $0.82
GPT-6.1 Terra 53.3% $1.55
GPT-6 Luna 53% $0.15
Muse Spark Open 53% $3.00
DeepSeek V4 Pro 0813 52.9% $0.88
MiniMax-M3 48.4% $1.22
Mercury 3 47.2% $0.10
Beam 1.0 46.5%
Nemotron 3.5 Super VL 41% $0.48
Inkling 38.3% $0.97
Inkling-Small 36.3% $0.18

2 models have held number one since its release in Oct 2026

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

Anthropic Google DeepMind
Own leaderboard 2 leaders
Gemini 4 Carbon 67%

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 Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon Qwen3.8 Max (0902) MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon GLM-5.3

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

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%