Benchmark Almanac

Vals Finance Agent v2

Gemini 4 Argon 65.4% leads since September 2026
Gemini 4 Argon 65.4% leads since September 2026
In use

927 expert-reviewed research questions on public-company filings. Partial credit only after required checks pass. Private test split.

Rebuilt with 927 harder questions in nine workflow categories, scored with partial credit. Scores do not compare with v1.1.

Made by
Vals AI
Released
May 2026
Kind
Finance
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Public, plus a private test set
Strongest model tested
GPT-6 Astra #3 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026

In the authors’ words

Artificial Intelligence (AI) technology has emerged as a transformative force in financial analysis and the finance industry, though significant questions remain about the full capabilities of Large Language Model (LLM) agents in this domain. We present the Finance Agent Benchmark, featuring challenging and diverse real-world finance research problems that require LLMs to perform complex analysis using recent SEC filings. We construct the benchmark using a taxonomy of nine financial task categories, developed in consultation with experts from banks, hedge funds, and private equity firms. The dataset includes 537 expert-authored questions covering tasks from information retrieval to complex financial modeling, each validated through a rigorous review process to ensure accuracy and relevance. Moreover, we implement an agentic harness that equips LLMs with tools sufficient to produce accurate responses, including Google Search and EDGAR database access. Overall, the Finance Agent Benchmark provides a comprehensive testbed for measuring the progress of LLM-driven finance agents. Our evaluation reveals significant limitations in current AI capabilities - even the best-performing model (OpenAI o3) achieved only 46.8% accuracy at an average cost of $3.79 per query. This underscores the need for further advancements before reliable deployment in high-stakes finance settings.

Abstract of Finance Agent Benchmark: Benchmarking LLMs on Real-world Financial Research Tasks (arXiv:2508.00828, CC0)
80
Models scored on its own leaderboard
10
Launch posts list it, from 6 labs
9
System cards report it, from 4 labs

Gemini 4 Argon leads with 65.4%

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

Gemini 4 Argon leads with 65.4%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

80 models. Scroll sideways for the rest.

  1. 65.4 Gemini 4 Argon
  2. 61.4 Gemini 3.8 Flash
  3. 61.3 Muse Spark Open
  4. 60.6 Muse Spark 1.2
  5. 60.0 Muse Spark 1.3
  6. 59.1 Muse Wonder Full
  7. 59.0 Gemini 3.7 Flash
  8. 58.9 Claude Fable 5.1
  9. 58.6 Claude Opus 5
  10. 58.6 Claude Opus 5.5
  11. 58.1 Claude Sonnet 5.5
  12. 57.9 Qwen3.8 Max (0902)
  13. 57.9 Gemini 3.5 Flash
  14. 57.9 GLM-5.3-Flash
  15. 57.3 MiMo-V2.6 Pro
  16. 57.2 Muse Spark 1.1
  17. 57.0 DeepSeek V4.1 Flash 1017
  18. 56.6 Muse Wonder
  19. 56.3 Claude Fable 5
  20. 56.3 Gemini 3.6 Flash
  21. 56.3 MiMo-V2.6 Flash
  22. 55.8 GLM-5.3
  23. 55.1 Tencent Hy4 preview
  24. 55.0 GPT-5.6 Luna
  25. 54.9 Ling-3.0-flash-Fin
  26. 54.7 Mistral Large 4 Preview
  27. 54.4 GPT-5.6 Terra
  28. 54.1 Claude Haiku 5.5
  29. 53.9 Claude Opus 4.8
  30. 53.9 Claude Sonnet 5
  31. 53.8 GPT-5.6 Sol
  32. 53.7 Grok 4.6
  33. 53.5 GPT-6 Astra
  34. 53.5 DeepSeek V4.1 Flash
  35. 53.1 Kimi K3
  36. 52.3 Grok 4.7
  37. 52.0 GPT-6.1 Sol
  38. 51.9 Ember-1
  39. 51.8 GPT-5.5
  40. 51.5 Claude Opus 4.7
  41. 51.0 Claude Sonnet 4.6
  42. 50.7 Step 5 Preview
  43. 50.6 Qwen 3.8 Max
  44. 50.4 DeepSeek V4 Pro 0813
  45. 49.9 GPT-6 Luna
  46. 49.7 GLM-5.2
  47. 49.5 DeepSeek V4 Flash 0731
  48. 49.0 GPT-6 Sol
  49. 48.5 Qwen 3.8 27B
  50. 48.4 Grok 4.5
  51. 48.3 MiniMax-M3
  52. 47.8 Qwen3.7-Max
  53. 47.4 Gemini 3.5 Flash-Lite
  54. 46.6 Inkling
  55. 45.4 GPT-5.4 Mini
  56. 44.9 Kimi K2.6
  57. 44.8 GLM-5.1
  58. 44.1 DeepSeek-V4-Pro
  59. 43.0 Gemini 3.1 Pro
  60. 42.5 Gemini 3 Flash
  61. 41.5 MiMo-V2.5-Pro
  62. 41.3 Inkling-Small
  63. 40.9 Qwen 3.6 Plus
  64. 38.2 GPT-5.4 Nano
  65. 38.2 Qwen3.7-Plus
  66. 37.7 Grok 4.3
  67. 37.7 Nemotron 3 Ultra
  68. 36.7 MiMo-V2.5
  69. 35.8 Kimi K2.5
  70. 32.1 Mistral Medium 3.5
  71. 31.0 Claude Haiku 4.5
  72. 30.3 Ling-3.0-flash
  73. 30.0 Gemini 3.1 Flash-Lite
  74. 28.5 Grok 4.20
  75. 27.9 MiniMax-M2.7
  76. 25.0 Laguna M.1
  77. 18.9 Mercury 2.5
  78. 18.5 Nemotron 3.5 Lightning 30B-A3B
  79. 15.6 Laguna XS.2
  80. 9.04 Command A+
  1. 1 Gemini 4 Argon 65.4%
  2. 2 Gemini 3.8 Flash 61.4%
  3. 3 Muse Spark Open 61.3%
  4. 4 Muse Spark 1.2 60.6%
  5. 5 Muse Spark 1.3 max 60%
  6. 6 Muse Wonder Full 59.1%
  7. 7 Gemini 3.7 Flash 59%
  8. 8 Claude Fable 5.1 58.9%
  9. 9 Claude Opus 5 58.6%
  10. 10 Claude Opus 5.5 58.6%
  11. 11 Claude Sonnet 5.5 58.1%
  12. 12 Qwen3.8 Max (0902) 57.9%
  13. 13 Gemini 3.5 Flash 57.9%
  14. 14 GLM-5.3-Flash 57.9%
  15. 15 MiMo-V2.6 Pro 57.3%
  16. 16 Muse Spark 1.1 57.2%
  17. 17 DeepSeek V4.1 Flash 1017 57%
  18. 18 Muse Wonder 56.6%
  19. 19 Claude Fable 5 56.3%
  20. 20 Gemini 3.6 Flash 56.3%
  21. 21 MiMo-V2.6 Flash 56.3%
  22. 22 GLM-5.3 55.8%
  23. 23 Tencent Hy4 preview 55.1%
  24. 24 GPT-5.6 Luna 55%
  25. 25 Ling-3.0-flash-Fin 54.9%
  26. 26 Mistral Large 4 Preview 54.7%
  27. 27 GPT-5.6 Terra 54.4%
  28. 28 Claude Haiku 5.5 54.1%
  29. 29 Claude Opus 4.8 53.9%
  30. 30 Claude Sonnet 5 53.9%
  31. 31 GPT-5.6 Sol 53.8%
  32. 32 Grok 4.6 53.7%
  33. 33 GPT-6 Astra 53.5%
  34. 34 DeepSeek V4.1 Flash 53.5%
  35. 35 Kimi K3 53.1%
  36. 36 Grok 4.7 52.3%
  37. 37 GPT-6.1 Sol 52%
  38. 38 Ember-1 51.9%
  39. 39 GPT-5.5 51.8%
  40. 40 Claude Opus 4.7 51.5%
  41. 41 Claude Sonnet 4.6 51%
  42. 42 Step 5 Preview 50.7%
  43. 43 Qwen 3.8 Max 50.6%
  44. 44 DeepSeek V4 Pro 0813 50.4%
  45. 45 GPT-6 Luna 49.9%
  46. 46 GLM-5.2 49.7%
  47. 47 DeepSeek V4 Flash 0731 49.5%
  48. 48 GPT-6 Sol 49%
  49. 49 Qwen 3.8 27B 48.5%
  50. 50 Grok 4.5 48.4%
  51. 51 MiniMax-M3 48.3%
  52. 52 Qwen3.7-Max 47.8%
  53. 53 Gemini 3.5 Flash-Lite 47.4%
  54. 54 Inkling 46.6%
  55. 55 GPT-5.4 Mini 45.4%
  56. 56 Kimi K2.6 44.9%
  57. 57 GLM-5.1 44.8%
  58. 58 DeepSeek-V4-Pro 44.1%
  59. 59 Gemini 3.1 Pro 43%
  60. 60 Gemini 3 Flash 42.5%
  61. 61 MiMo-V2.5-Pro 41.5%
  62. 62 Inkling-Small 41.3%
  63. 63 Qwen 3.6 Plus 40.9%
  64. 64 GPT-5.4 Nano 38.2%
  65. 65 Qwen3.7-Plus 38.2%
  66. 66 Grok 4.3 37.7%
  67. 67 Nemotron 3 Ultra 37.7%
  68. 68 MiMo-V2.5 36.7%
  69. 69 Kimi K2.5 thinking 35.8%
  70. 70 Mistral Medium 3.5 32.1%
  71. 71 Claude Haiku 4.5 thinking 31%
  72. 72 Ling-3.0-flash 30.3%
  73. 73 Gemini 3.1 Flash-Lite 30%
  74. 74 Grok 4.20 reasoning 28.5%
  75. 75 MiniMax-M2.7 27.9%
  76. 76 Laguna M.1 25%
  77. 77 Mercury 2.5 18.9%
  78. 78 Nemotron 3.5 Lightning 30B-A3B 18.5%
  79. 79 Laguna XS.2 15.6%
  80. 80 Command A+ 9.04%

Source: Official Vals Finance Agent v2 leaderboard

82 models. Scroll sideways for the rest.

  1. 65.4 Gemini 4 Argon
  2. 61.4 Gemini 3.8 Flash
  3. 61.3 Muse Spark Open
  4. 60.6 Muse Spark 1.2
  5. 60.0 Muse Spark 1.3
  6. 59.1 Muse Wonder Full
  7. 59.0 Gemini 3.7 Flash
  8. 58.9 Claude Fable 5.1
  9. 58.6 Claude Opus 5
  10. 58.6 Claude Opus 5.5
  11. 58.1 Claude Sonnet 5.5
  12. 57.9 Gemini 3.5 Flash
  13. 57.9 Qwen3.8 Max (0902)
  14. 57.9 Ling-3.1-flash
  15. 57.9 GLM-5.3-Flash
  16. 57.3 MiMo-V2.6 Pro
  17. 57.2 Muse Spark 1.1
  18. 57.0 DeepSeek V4.1 Flash 1017
  19. 56.6 Muse Wonder
  20. 56.3 Claude Fable 5
  21. 56.3 Gemini 3.6 Flash
  22. 56.3 MiMo-V2.6 Flash
  23. 55.8 GLM-5.3
  24. 55.1 Tencent Hy4 preview
  25. 55.0 GPT-5.6 Luna
  26. 54.9 Ling-3.0-flash-Fin
  27. 54.7 Mistral Large 4
  28. 54.7 Mistral Large 4 Preview
  29. 54.4 GPT-5.6 Terra
  30. 54.4 Kimi K3
  31. 54.1 Claude Haiku 5.5
  32. 53.9 Claude Opus 4.8
  33. 53.9 Claude Sonnet 5
  34. 53.8 GPT-5.6 Sol
  35. 53.7 Grok 4.6
  36. 53.5 GPT-6 Astra
  37. 53.5 DeepSeek V4.1 Flash
  38. 52.3 Grok 4.7
  39. 52.0 GPT-6.1 Sol
  40. 51.9 Ember-1
  41. 51.8 GPT-5.5
  42. 51.5 Claude Opus 4.7
  43. 51.0 Claude Sonnet 4.6
  44. 50.7 Step 5 Preview
  45. 50.6 Qwen 3.8 Max
  46. 50.4 DeepSeek V4 Pro 0813
  47. 49.9 GPT-6 Luna
  48. 49.7 GLM-5.2
  49. 49.5 DeepSeek V4 Flash 0731
  50. 49.0 GPT-6 Sol
  51. 48.5 Qwen 3.8 27B
  52. 48.4 Grok 4.5
  53. 48.3 MiniMax-M3
  54. 47.8 Qwen3.7-Max
  55. 47.4 Gemini 3.5 Flash-Lite
  56. 46.6 Inkling
  57. 45.4 GPT-5.4 Mini
  58. 44.9 Kimi K2.6
  59. 44.8 GLM-5.1
  60. 44.1 DeepSeek-V4-Pro
  61. 43.0 Gemini 3.1 Pro
  62. 42.5 Gemini 3 Flash
  63. 41.5 MiMo-V2.5-Pro
  64. 41.3 Inkling-Small
  65. 40.9 Qwen 3.6 Plus
  66. 38.2 GPT-5.4 Nano
  67. 38.2 Qwen3.7-Plus
  68. 37.7 Grok 4.3
  69. 37.7 Nemotron 3 Ultra
  70. 36.7 MiMo-V2.5
  71. 35.8 Kimi K2.5
  72. 32.1 Mistral Medium 3.5
  73. 31.0 Claude Haiku 4.5
  74. 30.3 Ling-3.0-flash
  75. 30.0 Gemini 3.1 Flash-Lite
  76. 28.5 Grok 4.20
  77. 27.9 MiniMax-M2.7
  78. 25.0 Laguna M.1
  79. 18.9 Mercury 2.5
  80. 18.5 Nemotron 3.5 Lightning 30B-A3B
  81. 15.6 Laguna XS.2
  82. 9.04 Command A+
  1. 1 Gemini 4 Argon 65.4%
  2. 2 Gemini 3.8 Flash 61.4%
  3. 3 Muse Spark Open 61.3%
  4. 4 Muse Spark 1.2 60.6%
  5. 5 Muse Spark 1.3 max 60%
  6. 6 Muse Wonder Full 59.1%
  7. 7 Gemini 3.7 Flash 59%
  8. 8 Claude Fable 5.1 58.9%
  9. 9 Claude Opus 5 58.6%
  10. 10 Claude Opus 5.5 58.6%
  11. 11 Claude Sonnet 5.5 58.1%
  12. 12 Gemini 3.5 Flash 57.9%
  13. 13 Qwen3.8 Max (0902) 57.9%
  14. 14 Ling-3.1-flash 57.9%
  15. 15 GLM-5.3-Flash 57.9%
  16. 16 MiMo-V2.6 Pro 57.3%
  17. 17 Muse Spark 1.1 57.2%
  18. 18 DeepSeek V4.1 Flash 1017 57%
  19. 19 Muse Wonder 56.6%
  20. 20 Claude Fable 5 56.3%
  21. 21 Gemini 3.6 Flash 56.3%
  22. 22 MiMo-V2.6 Flash 56.3%
  23. 23 GLM-5.3 55.8%
  24. 24 Tencent Hy4 preview 55.1%
  25. 25 GPT-5.6 Luna 55%
  26. 26 Ling-3.0-flash-Fin 54.9%
  27. 27 Mistral Large 4 54.7%
  28. 28 Mistral Large 4 Preview 54.7%
  29. 29 GPT-5.6 Terra 54.4%
  30. 30 Kimi K3 54.4%
  31. 31 Claude Haiku 5.5 54.1%
  32. 32 Claude Opus 4.8 53.9%
  33. 33 Claude Sonnet 5 53.9%
  34. 34 GPT-5.6 Sol 53.8%
  35. 35 Grok 4.6 53.7%
  36. 36 GPT-6 Astra 53.5%
  37. 37 DeepSeek V4.1 Flash 53.5%
  38. 38 Grok 4.7 52.3%
  39. 39 GPT-6.1 Sol 52%
  40. 40 Ember-1 51.9%
  41. 41 GPT-5.5 51.8%
  42. 42 Claude Opus 4.7 51.5%
  43. 43 Claude Sonnet 4.6 51%
  44. 44 Step 5 Preview 50.7%
  45. 45 Qwen 3.8 Max 50.6%
  46. 46 DeepSeek V4 Pro 0813 50.4%
  47. 47 GPT-6 Luna 49.9%
  48. 48 GLM-5.2 49.7%
  49. 49 DeepSeek V4 Flash 0731 49.5%
  50. 50 GPT-6 Sol 49%
  51. 51 Qwen 3.8 27B 48.5%
  52. 52 Grok 4.5 48.4%
  53. 53 MiniMax-M3 48.3%
  54. 54 Qwen3.7-Max 47.8%
  55. 55 Gemini 3.5 Flash-Lite 47.4%
  56. 56 Inkling 46.6%
  57. 57 GPT-5.4 Mini 45.4%
  58. 58 Kimi K2.6 44.9%
  59. 59 GLM-5.1 44.8%
  60. 60 DeepSeek-V4-Pro 44.1%
  61. 61 Gemini 3.1 Pro 43%
  62. 62 Gemini 3 Flash 42.5%
  63. 63 MiMo-V2.5-Pro 41.5%
  64. 64 Inkling-Small 41.3%
  65. 65 Qwen 3.6 Plus 40.9%
  66. 66 GPT-5.4 Nano 38.2%
  67. 67 Qwen3.7-Plus 38.2%
  68. 68 Grok 4.3 37.7%
  69. 69 Nemotron 3 Ultra 37.7%
  70. 70 MiMo-V2.5 36.7%
  71. 71 Kimi K2.5 thinking 35.8%
  72. 72 Mistral Medium 3.5 32.1%
  73. 73 Claude Haiku 4.5 thinking 31%
  74. 74 Ling-3.0-flash 30.3%
  75. 75 Gemini 3.1 Flash-Lite 30%
  76. 76 Grok 4.20 reasoning 28.5%
  77. 77 MiniMax-M2.7 27.9%
  78. 78 Laguna M.1 25%
  79. 79 Mercury 2.5 18.9%
  80. 80 Nemotron 3.5 Lightning 30B-A3B 18.5%
  81. 81 Laguna XS.2 15.6%
  82. 82 Command A+ 9.04%

Sources: Official Vals Finance Agent v2 leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 Gemini 4 Argon 65.4% Sep 2026 Official leaderboard
2 Gemini 3.8 Flash 61.4% Sep 2026 Official leaderboard
3 Muse Spark Open 61.3% Oct 2026 Official leaderboard
4 Muse Spark 1.2 60.6% Aug 2026 Official leaderboard
5 Muse Spark 1.3 60% max Sep 2026 Official leaderboard
6 Muse Wonder Full 59.1% Oct 2026 Official leaderboard
7 Gemini 3.7 Flash 59% Aug 2026 Official leaderboard
8 Claude Fable 5.1 58.9% Sep 2026 Official leaderboard
9 Claude Opus 5 58.6% Jul 2026 Official leaderboard
10 Claude Opus 5.5 58.6% Sep 2026 Official leaderboard
11 Claude Sonnet 5.5 58.1% Sep 2026 Official leaderboard
12 Qwen3.8 Max (0902) 57.9% Sep 2026 Official leaderboard
13 Gemini 3.5 Flash 57.9% May 2026 Official leaderboard
14 GLM-5.3-Flash 57.9% Aug 2026 Official leaderboard
15 MiMo-V2.6 Pro 57.3% Sep 2026 Official leaderboard
16 Muse Spark 1.1 57.2% Jul 2026 Official leaderboard
17 DeepSeek V4.1 Flash 1017 57% Oct 2026 Official leaderboard
18 Muse Wonder 56.6% Oct 2026 Official leaderboard
19 Claude Fable 5 56.3% Jun 2026 Official leaderboard
20 Gemini 3.6 Flash 56.3% Jul 2026 Official leaderboard
21 MiMo-V2.6 Flash 56.3% Sep 2026 Official leaderboard
22 GLM-5.3 55.8% Aug 2026 Official leaderboard
23 Tencent Hy4 preview 55.1% Aug 2026 Official leaderboard
24 GPT-5.6 Luna 55% Jul 2026 Official leaderboard
25 Ling-3.0-flash-Fin 54.9% Sep 2026 Official leaderboard
26 Mistral Large 4 Preview 54.7% Oct 2026 Official leaderboard
27 GPT-5.6 Terra 54.4% Jul 2026 Official leaderboard
28 Claude Haiku 5.5 54.1% Oct 2026 Official leaderboard
29 Claude Opus 4.8 53.9% May 2026 Official leaderboard
30 Claude Sonnet 5 53.9% Jun 2026 Official leaderboard
31 GPT-5.6 Sol 53.8% Jul 2026 Official leaderboard
32 Grok 4.6 53.7% Aug 2026 Official leaderboard
33 GPT-6 Astra 53.5% Sep 2026 Official leaderboard
34 DeepSeek V4.1 Flash 53.5% Sep 2026 Official leaderboard
35 Kimi K3 53.1% Jul 2026 Official leaderboard
36 Grok 4.7 52.3% Sep 2026 Official leaderboard
37 GPT-6.1 Sol 52% Sep 2026 Official leaderboard
38 Ember-1 51.9% Sep 2026 Official leaderboard
39 GPT-5.5 51.8% Apr 2026 Official leaderboard
40 Claude Opus 4.7 51.5% Apr 2026 Official leaderboard
41 Claude Sonnet 4.6 51% Feb 2026 Official leaderboard
42 Step 5 Preview 50.7% Sep 2026 Official leaderboard
43 Qwen 3.8 Max 50.6% Aug 2026 Official leaderboard
44 DeepSeek V4 Pro 0813 50.4% Aug 2026 Official leaderboard
45 GPT-6 Luna 49.9% Sep 2026 Official leaderboard
46 GLM-5.2 49.7% Jun 2026 Official leaderboard
47 DeepSeek V4 Flash 0731 49.5% Jul 2026 Official leaderboard
48 GPT-6 Sol 49% Sep 2026 Official leaderboard
49 Qwen 3.8 27B 48.5% Aug 2026 Official leaderboard
50 Grok 4.5 48.4% Jul 2026 Official leaderboard
51 MiniMax-M3 48.3% Jun 2026 Official leaderboard
52 Qwen3.7-Max 47.8% May 2026 Official leaderboard
53 Gemini 3.5 Flash-Lite 47.4% Jul 2026 Official leaderboard
54 Inkling 46.6% Jul 2026 Official leaderboard
55 GPT-5.4 Mini 45.4% Mar 2026 Official leaderboard
56 Kimi K2.6 44.9% Apr 2026 Official leaderboard
57 GLM-5.1 44.8% Apr 2026 Official leaderboard
58 DeepSeek-V4-Pro 44.1% Apr 2026 Official leaderboard
59 Gemini 3.1 Pro 43% Feb 2026 Official leaderboard
60 Gemini 3 Flash 42.5% Dec 2025 Official leaderboard
61 MiMo-V2.5-Pro 41.5% Apr 2026 Official leaderboard
62 Inkling-Small 41.3% Jul 2026 Official leaderboard
63 Qwen 3.6 Plus 40.9% Mar 2026 Official leaderboard
64 GPT-5.4 Nano 38.2% Mar 2026 Official leaderboard
65 Qwen3.7-Plus 38.2% Jun 2026 Official leaderboard
66 Grok 4.3 37.7% Apr 2026 Official leaderboard
67 Nemotron 3 Ultra 37.7% Jun 2026 Official leaderboard
68 MiMo-V2.5 36.7% Apr 2026 Official leaderboard
69 Kimi K2.5 35.8% thinking Jan 2026 Official leaderboard
70 Mistral Medium 3.5 32.1% Apr 2026 Official leaderboard
71 Claude Haiku 4.5 31% thinking Oct 2025 Official leaderboard
72 Ling-3.0-flash 30.3% Aug 2026 Official leaderboard
73 Gemini 3.1 Flash-Lite 30% Mar 2026 Official leaderboard
74 Grok 4.20 28.5% reasoning Feb 2026 Official leaderboard
75 MiniMax-M2.7 27.9% Mar 2026 Official leaderboard
76 Laguna M.1 25% Apr 2026 Official leaderboard
77 Mercury 2.5 18.9% Sep 2026 Official leaderboard
78 Nemotron 3.5 Lightning 30B-A3B 18.5% Aug 2026 Official leaderboard
79 Laguna XS.2 15.6% Apr 2026 Official leaderboard
80 Command A+ 9.04% May 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 4 Argon 65.4% Sep 2026 Official leaderboard
2 Gemini 3.8 Flash 61.4% Sep 2026 Official leaderboard
3 Muse Spark Open 61.3% Oct 2026 Official leaderboard
4 Muse Spark 1.2 60.6% Aug 2026 Official leaderboard
5 Muse Spark 1.3 60% max Sep 2026 Official leaderboard
6 Muse Wonder Full 59.1% Oct 2026 Lab model card
7 Gemini 3.7 Flash 59% Aug 2026 Official leaderboard
8 Claude Fable 5.1 58.9% Sep 2026 Official leaderboard
9 Claude Opus 5 58.6% Jul 2026 Official leaderboard
10 Claude Opus 5.5 58.6% Sep 2026 Official leaderboard
11 Claude Sonnet 5.5 58.1% Sep 2026 Official leaderboard
12 Gemini 3.5 Flash 57.9% May 2026 Lab model card
13 Qwen3.8 Max (0902) 57.9% Sep 2026 Official leaderboard
14 Ling-3.1-flash 57.9% Oct 2026 Lab model card
15 GLM-5.3-Flash 57.9% Aug 2026 Official leaderboard
16 MiMo-V2.6 Pro 57.3% Sep 2026 Official leaderboard
17 Muse Spark 1.1 57.2% Jul 2026 Official leaderboard
18 DeepSeek V4.1 Flash 1017 57% Oct 2026 Official leaderboard
19 Muse Wonder 56.6% Oct 2026 Official leaderboard
20 Claude Fable 5 56.3% Jun 2026 Official leaderboard
21 Gemini 3.6 Flash 56.3% Jul 2026 Official leaderboard
22 MiMo-V2.6 Flash 56.3% Sep 2026 Official leaderboard
23 GLM-5.3 55.8% Aug 2026 Official leaderboard
24 Tencent Hy4 preview 55.1% Aug 2026 Official leaderboard
25 GPT-5.6 Luna 55% Jul 2026 Official leaderboard
26 Ling-3.0-flash-Fin 54.9% Sep 2026 Official leaderboard
27 Mistral Large 4 54.7% Oct 2026 Lab model card
28 Mistral Large 4 Preview 54.7% Oct 2026 Official leaderboard
29 GPT-5.6 Terra 54.4% Jul 2026 Official leaderboard
30 Kimi K3 54.4% Jul 2026 Lab model card
31 Claude Haiku 5.5 54.1% Oct 2026 Official leaderboard
32 Claude Opus 4.8 53.9% May 2026 Official leaderboard
33 Claude Sonnet 5 53.9% Jun 2026 Official leaderboard
34 GPT-5.6 Sol 53.8% Jul 2026 Official leaderboard
35 Grok 4.6 53.7% Aug 2026 Official leaderboard
36 GPT-6 Astra 53.5% Sep 2026 Official leaderboard
37 DeepSeek V4.1 Flash 53.5% Sep 2026 Official leaderboard
38 Grok 4.7 52.3% Sep 2026 Official leaderboard
39 GPT-6.1 Sol 52% Sep 2026 Official leaderboard
40 Ember-1 51.9% Sep 2026 Official leaderboard
41 GPT-5.5 51.8% Apr 2026 Official leaderboard
42 Claude Opus 4.7 51.5% Apr 2026 Official leaderboard
43 Claude Sonnet 4.6 51% Feb 2026 Official leaderboard
44 Step 5 Preview 50.7% Sep 2026 Official leaderboard
45 Qwen 3.8 Max 50.6% Aug 2026 Official leaderboard
46 DeepSeek V4 Pro 0813 50.4% Aug 2026 Official leaderboard
47 GPT-6 Luna 49.9% Sep 2026 Official leaderboard
48 GLM-5.2 49.7% Jun 2026 Official leaderboard
49 DeepSeek V4 Flash 0731 49.5% Jul 2026 Official leaderboard
50 GPT-6 Sol 49% Sep 2026 Official leaderboard
51 Qwen 3.8 27B 48.5% Aug 2026 Official leaderboard
52 Grok 4.5 48.4% Jul 2026 Official leaderboard
53 MiniMax-M3 48.3% Jun 2026 Official leaderboard
54 Qwen3.7-Max 47.8% May 2026 Official leaderboard
55 Gemini 3.5 Flash-Lite 47.4% Jul 2026 Official leaderboard
56 Inkling 46.6% Jul 2026 Official leaderboard
57 GPT-5.4 Mini 45.4% Mar 2026 Official leaderboard
58 Kimi K2.6 44.9% Apr 2026 Official leaderboard
59 GLM-5.1 44.8% Apr 2026 Official leaderboard
60 DeepSeek-V4-Pro 44.1% Apr 2026 Official leaderboard
61 Gemini 3.1 Pro 43% Feb 2026 Official leaderboard
62 Gemini 3 Flash 42.5% Dec 2025 Official leaderboard
63 MiMo-V2.5-Pro 41.5% Apr 2026 Official leaderboard
64 Inkling-Small 41.3% Jul 2026 Official leaderboard
65 Qwen 3.6 Plus 40.9% Mar 2026 Official leaderboard
66 GPT-5.4 Nano 38.2% Mar 2026 Official leaderboard
67 Qwen3.7-Plus 38.2% Jun 2026 Official leaderboard
68 Grok 4.3 37.7% Apr 2026 Official leaderboard
69 Nemotron 3 Ultra 37.7% Jun 2026 Official leaderboard
70 MiMo-V2.5 36.7% Apr 2026 Official leaderboard
71 Kimi K2.5 35.8% thinking Jan 2026 Official leaderboard
72 Mistral Medium 3.5 32.1% Apr 2026 Official leaderboard
73 Claude Haiku 4.5 31% thinking Oct 2025 Official leaderboard
74 Ling-3.0-flash 30.3% Aug 2026 Official leaderboard
75 Gemini 3.1 Flash-Lite 30% Mar 2026 Official leaderboard
76 Grok 4.20 28.5% reasoning Feb 2026 Official leaderboard
77 MiniMax-M2.7 27.9% Mar 2026 Official leaderboard
78 Laguna M.1 25% Apr 2026 Official leaderboard
79 Mercury 2.5 18.9% Sep 2026 Official leaderboard
80 Nemotron 3.5 Lightning 30B-A3B 18.5% Aug 2026 Official leaderboard
81 Laguna XS.2 15.6% Apr 2026 Official leaderboard
82 Command A+ 9.04% May 2026 Official leaderboard

Labs report scores for 11 models in their cards. Switch the scores above to rank them together.

The best score, 65.4%, costs $4.38 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Spark 1.2 comes within 4.8 points of the best for $0.77 a task, 6 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

NVIDIA Z.ai Google DeepMind Meta Ant Group Xiaomi Corp Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Argon Gemini 3.8 Flash Muse Spark Open GLM-5.3-Flash Ling-3.0-flash-Fin Nemotron 3.5 Lightning 30B-A3B 0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Argon Muse Spark 1.3, max

Source: Official Vals Finance Agent v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon 65.4% $4.38 70,009
Gemini 3.8 Flash 61.4% $2.00 57,576
Muse Spark Open 61.3% $0.77 24,503
Muse Spark 1.2 60.6% $0.77 27,152
Muse Spark 1.3 max 60% $0.76 36,075
Muse Wonder Full 59.1% $2.39 40,907
Gemini 3.7 Flash 59% $1.48 37,527
Muse Spark 1.3 58.9% $0.74 32,330
Claude Fable 5.1 58.9% $8.35 73,400
Claude Opus 5 58.6% $5.12 50,936
Claude Opus 5.5 58.6% $9.03 245,444
Claude Sonnet 5.5 58.1% $6.39 309,344
Qwen3.8 Max (0902) 57.9% $3.03 157,939
Gemini 3.5 Flash 57.9% $2.51 59,518
GLM-5.3-Flash 57.9% $0.10 59,797
MiMo-V2.6 Pro 57.3% $0.20 29,729
Muse Spark 1.1 57.2% $0.72 33,954
DeepSeek V4.1 Flash 1017 57% $0.46 196,526
Muse Wonder 56.6% $2.67 35,481
Claude Fable 5 56.3% $8.06 44,377
Gemini 3.6 Flash 56.3% $1.40 36,216
MiMo-V2.6 Flash 56.3% $0.07 33,508
GLM-5.3 55.8% $1.07 81,621
Tencent Hy4 preview 55.1% $0.59 92,415
GPT-5.6 Luna 55% $0.28 116,832
Ling-3.0-flash-Fin 54.9% $0.04 43,134
Mistral Large 4 Preview 54.7% $0.61 61,943
GPT-5.6 Terra 54.4% $3.63 150,600
Claude Haiku 5.5 54.1% $1.64 462,161
Claude Opus 4.8 53.9% $4.22 38,447
Claude Sonnet 5 53.9% $0.75 22,449
GPT-5.6 Sol 53.8% $1.25 13,815
Grok 4.6 53.7% $1.66 54,559
GPT-6 Astra 53.5% $6.82 49,773
DeepSeek V4.1 Flash 53.5% $0.21 75,050
Kimi K3 53.1% $1.91 36,255
Grok 4.7 52.3% $2.72 52,063
GPT-6.1 Sol 52% $1.62 49,553
Ember-1 51.9% $1.95 29,994
GPT-5.5 51.8% $4.15 39,544
Claude Opus 4.7 51.5% $4.03 27,319
Claude Sonnet 4.6 51% $2.41 44,830
Step 5 Preview 50.7% $0.62 46,157
Qwen 3.8 Max 50.6% $1.24 63,675
DeepSeek V4 Pro 0813 50.4% $0.88 68,049
GPT-6 Luna 49.9% $0.12 101,625
GLM-5.2 49.7% $0.71 27,074
DeepSeek V4 Flash 0731 49.5% $0.28 60,941
GPT-6 Sol 49% $2.12 50,548
Qwen 3.8 27B 48.5% $0.75 93,068
Grok 4.5 48.4% $1.16 28,908
MiniMax-M3 48.3% $0.32 22,377
Qwen3.7-Max 47.8% $0.55 8,885
Gemini 3.5 Flash-Lite 47.4% $0.39 24,223
Inkling 46.6% $1.00 55,359
GPT-5.4 Mini 45.4% $1.20 162,092
Kimi K2.6 44.9% $1.14 52,313
GLM-5.1 44.8% $0.62 35,358
DeepSeek-V4-Pro 44.1% $0.88 49,019
Gemini 3.1 Pro 43% $1.59 17,179
Gemini 3 Flash 42.5% $0.57 33,818
MiMo-V2.5-Pro 41.5% $0.21 19,405
Inkling-Small 41.3% $0.23 62,274
Qwen 3.6 Plus 40.9% $0.81 25,710
GPT-5.4 Nano 38.2% $0.16 23,642
Qwen3.7-Plus 38.2% $0.36 22,216
Grok 4.3 37.7% $0.85 34,520
Nemotron 3 Ultra 37.7% 17,123
MiMo-V2.5 36.7% $0.09 18,757
Kimi K2.5 thinking 35.8% $0.29 17,402
Mistral Medium 3.5 32.1% $1.91 54,334
Claude Haiku 4.5 thinking 31% $0.60 17,743
Ling-3.0-flash 30.3% $0.05 29,301
Gemini 3.1 Flash-Lite 30% $0.14 10,041
Grok 4.20 reasoning 28.5% $0.79 27,426
MiniMax-M2.7 27.9% $0.18 35,240
Laguna M.1 25%
Mercury 2.5 18.9% $0.11 12,959
Nemotron 3.5 Lightning 30B-A3B 18.5% $0.03 42,838
Laguna XS.2 15.6% 19,642
Command A+ 9.04% 45,109

6 models have held number one on the official leaderboard since its release in May 2026 6 models have held number one with claims included since its release in May 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

OpenAI Google DeepMind Anthropic Meta
Own leaderboard 6 leaders
Gemini 4 Argon 65.4%
Lab claims 3 leaders
Gemini 4 Argon 65.4%

Sources: Official Vals Finance Agent v2 leaderboard and labs’ model cards

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% 2026 Near the ceiling (85%) Gemini 4 Argon Muse Spark 1.2 GPT-5.5 Gemini 3 Flash Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Opus 5

Sources: Official Vals Finance Agent v2 leaderboard and labs’ model cards

5 labs 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
Anthropic
Meta
Moonshot AI
Mistral AI
Ant Group
DateLaunch postScoreAs the post writes it
Oct 13, 2026 Ling-3.1-flash 57.9 Finance Agent V2
Oct 13, 2026 Gemini 4 Argon 65.4 Vals Finance Agent v2
Oct 12, 2026 Muse Wonder 56.6 Vals Finance Agent V2
Oct 12, 2026 Muse Wonder Full 59.1 Vals Finance Agent V2
Oct 6, 2026 Mistral Large 4 54.7 Finance Agent: Vals.ai - Finance Agent V2
Sep 2, 2026 Gemini 3.8 Flash 61.4 Vals Finance Agent v2
Jul 16, 2026 Kimi K3 54.4 Finance Agent v2
Jul 9, 2026 Muse Spark 1.1 57.2 Finance Agent v2 (Agentic financial anaysis)
May 28, 2026 Claude Opus 4.8 53.9 Agentic financial analysis Finance Agent v2
May 19, 2026 Gemini 3.5 Flash 57.9 Finance Agent v2

9 system cards report it, from 4 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 13, 2026 Gemini 4 Argon model card 65.4 Vals Finance Agent v2 (results table, p.5) Capabilities
Oct 12, 2026 Muse Wonder model card 56.6 Vals Finance Agent V2, private test set accuracy from Vals AI, Muse Wonder (max) (capabilities table image, page 3) Capabilities
Oct 12, 2026 Muse Wonder Full model card 59.1 Vals Finance Agent V2, private test set accuracy from Vals AI, Muse Wonder Full (max) (capabilities table image, page 3) Capabilities
Sep 2, 2026 Gemini 3.8 Flash model card 61.4 Vals Finance Agent v2 (results table, p.5) Capabilities
Jul 27, 2026 Kimi K3 technical report 54.4 Finance Agent v2, cited from Vals AI (Table 2) Capabilities
Jul 9, 2026 Muse Spark 1.1 system card 57.2 Finance Agent v2 (Figure 44) Capabilities
Jun 9, 2026 Claude Fable 5 system card 56.3 Finance Agent Benchmark v2, run by Vals AI, adaptive thinking at max effort (8.17.2) Capabilities
May 28, 2026 Claude Opus 4.8 system card 53.9 Finance Agent v2, run by Vals AI at max effort, 53.92% in text (Table 8.1.A, 8.13.2) Capabilities
May 19, 2026 Gemini 3.5 Flash model card 57.9 Finance Agent v2 (results table, p.4) Capabilities

9 lab claims next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 8 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Kimi K3 54.4% Finance Agent v2 53.1% Leaderboard +1.3 pts
Gemini 3.5 Flash 57.9% Finance Agent v2 57.9% Leaderboard Same
Gemini 3.8 Flash 61.4% Vals Finance Agent v2 61.4% Leaderboard Same
Muse Wonder Full 59.1% Vals Finance Agent V2 59.1% Leaderboard Same
Muse Wonder 56.6% Vals Finance Agent V2 56.6% Leaderboard Same
Claude Opus 4.8 53.9% Agentic financial analysis Finance Agent v2 53.9% Leaderboard Same
Muse Spark 1.1 57.2% Finance Agent v2 (Agentic financial anaysis) 57.2% Leaderboard Same
Gemini 4 Argon 65.4% Vals Finance Agent v2 65.4% Leaderboard Same
Claude Fable 5 56.3% Finance Agent Benchmark v2, run by Vals AI, adaptive thinking at max effort (8.17.2) 56.3% Leaderboard Same

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI Google DeepMind Meta
Vals Finance Agent v2.1 18 leaders
Claude Fable 5.2 69.5%
Vals Finance Agent v1 Replaced by Agent, Feb 2026
Claude Opus 4.6 60.6%
Vals Finance Agent Replaced by v2, May 2026
Claude Opus 4.7 64.4%
Vals Finance Agent v2 6 leaders
Gemini 4 Argon 65.4%
Vals Finance Agent v2.1 2 leaders
Claude Fable 5.2 69.5%

Source: Official leaderboards

Replaces Vals Finance Agent Replaced by Vals Finance Agent v2.1

More in Law, finance and medicine

All 179 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%