Benchmark Almanac

APEX-Agents

Kimi K3.1 Pro 67.2% leads since October 2026
Gemini 3 Flash 24% leads since January 2026
Gemini 4 Carbon 84.2% leads since October 2026
In use

480 long-horizon tasks in banking, consulting and law worlds with docs, sheets, email and chat. Rubric-graded, Pass@1.

Repo is Mercor's Archipelago harness that runs the dataset. Dataset license forbids training on it.

Made by
Mercor
Released
January 2026
Kind
Simulated companies
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Strongest model tested
Kimi K3.1 Pro #16 on the capability index
Last in a launch post
October 2026
Last model added
November 2026
Scores refreshed
November 12th, 2026

In the authors’ words

We introduce the AI Productivity Index for Agents (APEX-Agents), a benchmark for assessing whether AI agents can execute long-horizon, cross-application tasks created by investment banking analysts, management consultants, and corporate lawyers. APEX-Agents requires agents to navigate realistic work environments with files and tools. We test eight agents for the leaderboard using Pass@1. Gemini 3 Flash (Thinking=High) achieves the highest score of 24.0%, followed by GPT-5.2 (Thinking=High), Claude Opus 4.5 (Thinking=High), and Gemini 3 Pro (Thinking=High). We open source the APEX-Agents benchmark (n=480) with all prompts, rubrics, gold outputs, files, and metadata. We also open source Archipelago, our infrastructure for agent execution and evaluation.

Abstract of APEX-Agents (arXiv:2601.14242, CC0)
12
Models scored in lab cards
11
Launch posts list it, from 7 labs
5
System cards report it, from 5 labs

Kimi K3.1 Pro leads with 67.2%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

Gemini 3 Flash leads with 24% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.

Gemini 4 Carbon leads with 84.2%, 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.

  1. 67.2 Kimi K3.1 Pro
  2. 57.5 Grok 4.6
  3. 41.0 Kimi K3
  4. 37.8 Step 5 Preview
  5. 37.1 Tencent Hy4 preview
  6. 33.8 Seed2.1 Pro
  7. 33.5 Gemini 3.1 Pro
  8. 27.9 Kimi K2.6
  9. 27.7 MiniMax-M3
  10. 25.6 Tencent Hy3
  11. 18.7 Solar Pro 4
  12. 16.6 Solar Open 2
  1. 1 Kimi K3.1 Pro 67.2%
  2. 2 Grok 4.6 57.5%
  3. 3 Kimi K3 41%
  4. 4 Step 5 Preview 37.8%
  5. 5 Tencent Hy4 preview 37.1%
  6. 6 Seed2.1 Pro 33.8%
  7. 7 Gemini 3.1 Pro 33.5%
  8. 8 Kimi K2.6 27.9%
  9. 9 MiniMax-M3 27.7%
  10. 10 Tencent Hy3 25.6%
  11. 11 Solar Pro 4 18.7%
  12. 12 Solar Open 2 16.6%

Source: Labs’ model cards

Source: Official APEX-Agents leaderboard

75 models. Scroll sideways for the rest.

  1. 84.2 Gemini 4 Carbon
  2. 82.4 Gemini 4 Argon
  3. 80.1 Claude Fable 5.2
  4. 75.5 Claude Sonnet 5.5
  5. 73.5 Claude Opus 5.5
  6. 69.8 DeepSeek V4.1 Flash 1017
  7. 68.6 Claude Fable 5.1
  8. 68.3 GPT-6.1 Terra
  9. 67.8 Gemini 3.7 Flash
  10. 67.2 Kimi K3.1 Pro
  11. 66.1 Muse Wonder
  12. 65.8 Claude Opus 5
  13. 65.3 Grok 4.6
  14. 64.7 GPT-6 Astra
  15. 64.3 Gemini 3.8 Flash
  16. 64.3 Muse Ember
  17. 63.6 Claude Fable 5
  18. 63.3 Qwen 3.8 Max
  19. 60.0 GPT-6.1 Sol
  20. 58.6 Muse Spark 1.3
  21. 58.2 GPT-5.6 Terra
  22. 56.6 GLM-5.3
  23. 56.2 Grok 4.5
  24. 56.1 Claude Haiku 5.5
  25. 55.1 GPT-5.5
  26. 54.6 Grok 4.7
  27. 54.5 Claude Sonnet 5
  28. 54.4 MiMo-V2.6 Pro
  29. 54.3 GPT-6 Sol
  30. 52.8 GLM-5.3-Flash
  31. 52.4 GPT-5.4
  32. 51.6 Mistral Large 4 Preview
  33. 51.4 GPT-5.6 Sol
  34. 49.2 Claude Opus 4.7
  35. 48.9 Claude Opus 4.8
  36. 48.8 Beam 1.0
  37. 47.8 Kimi K3
  38. 47.5 Qwen 3.8 27B
  39. 47.3 DeepSeek V4 Pro 0813
  40. 46.9 Gemini 3.6 Flash
  41. 46.8 Inkling 2 Small
  42. 46.3 Claude Opus 4.6
  43. 45.4 Muse Spark 1.2
  44. 45.2 GLM-5.2
  45. 44.3 GPT-6 Luna
  46. 43.0 Claude Sonnet 4.6
  47. 43.0 GPT-5.6 Luna
  48. 40.9 GLM-5.1
  49. 39.5 DeepSeek V4.1 Flash
  50. 37.8 Step 5 Preview
  51. 37.7 MiniMax-M3
  52. 37.6 Kimi K2.7 Code
  53. 37.1 Tencent Hy4 preview
  54. 35.1 Gemini 3.1 Pro
  55. 33.8 Seed2.1 Pro
  56. 33.8 Inkling
  57. 33.4 Muse Spark Open
  58. 31.8 Muse Spark 1.1
  59. 29.3 Gemini 3.5 Flash-Lite
  60. 27.9 Kimi K2.6
  61. 27.5 Gemini 3.5 Flash
  62. 25.6 Tencent Hy3
  63. 24.9 Qwen3.5 397B-A17B
  64. 24.0 Gemini 3 Flash
  65. 23.0 GPT-5.2
  66. 22.7 Nemotron 3 Ultra
  67. 21.3 DeepSeek-V3.2
  68. 18.7 Solar Pro 4
  69. 18.4 Gemini 3 Pro
  70. 18.4 Claude Opus 4.5
  71. 18.3 GPT-5
  72. 16.6 Solar Open 2
  73. 15.2 Grok 4
  74. 4.70 gpt-oss-120b
  75. 4.00 Kimi K2 Thinking
  1. 1 Gemini 4 Carbon 84.2%
  2. 2 Gemini 4 Argon 82.4%
  3. 3 Claude Fable 5.2 80.1%
  4. 4 Claude Sonnet 5.5 max 75.5%
  5. 5 Claude Opus 5.5 max 73.5%
  6. 6 DeepSeek V4.1 Flash 1017 max 69.8%
  7. 7 Claude Fable 5.1 68.6%
  8. 8 GPT-6.1 Terra max 68.3%
  9. 9 Gemini 3.7 Flash 67.8%
  10. 10 Kimi K3.1 Pro 67.2%
  11. 11 Muse Wonder max 66.1%
  12. 12 Claude Opus 5 max 65.8%
  13. 13 Grok 4.6 65.3%
  14. 14 GPT-6 Astra unknown thinking 64.7%
  15. 15 Gemini 3.8 Flash 64.3%
  16. 16 Muse Ember 64.3%
  17. 17 Claude Fable 5 63.6%
  18. 18 Qwen 3.8 Max 63.3%
  19. 19 GPT-6.1 Sol max 60%
  20. 20 Muse Spark 1.3 58.6%
  21. 21 GPT-5.6 Terra max 58.2%
  22. 22 GLM-5.3 unknown thinking 56.6%
  23. 23 Grok 4.5 unknown thinking 56.2%
  24. 24 Claude Haiku 5.5 max 56.1%
  25. 25 GPT-5.5 unknown thinking 55.1%
  26. 26 Grok 4.7 54.6%
  27. 27 Claude Sonnet 5 unknown thinking 54.5%
  28. 28 MiMo-V2.6 Pro 54.4%
  29. 29 GPT-6 Sol max 54.3%
  30. 30 GLM-5.3-Flash 52.8%
  31. 31 GPT-5.4 unknown thinking 52.4%
  32. 32 Mistral Large 4 Preview high 51.6%
  33. 33 GPT-5.6 Sol pro, max 51.4%
  34. 34 Claude Opus 4.7 max 49.2%
  35. 35 Claude Opus 4.8 max 48.9%
  36. 36 Beam 1.0 max 48.8%
  37. 37 Kimi K3 47.8%
  38. 38 Qwen 3.8 27B 47.5%
  39. 39 DeepSeek V4 Pro 0813 47.3%
  40. 40 Gemini 3.6 Flash 46.9%
  41. 41 Inkling 2 Small 46.8%
  42. 42 Claude Opus 4.6 max 46.3%
  43. 43 Muse Spark 1.2 45.4%
  44. 44 GLM-5.2 45.2%
  45. 45 GPT-6 Luna max 44.3%
  46. 46 Claude Sonnet 4.6 high 43%
  47. 47 GPT-5.6 Luna unknown thinking 43%
  48. 48 GLM-5.1 40.9%
  49. 49 DeepSeek V4.1 Flash 39.5%
  50. 50 Step 5 Preview 37.8%
  51. 51 MiniMax-M3 37.7%
  52. 52 Kimi K2.7 Code 37.6%
  53. 53 Tencent Hy4 preview 37.1%
  54. 54 Gemini 3.1 Pro 35.1%
  55. 55 Seed2.1 Pro 33.8%
  56. 56 Inkling 33.8%
  57. 57 Muse Spark Open 33.4%
  58. 58 Muse Spark 1.1 31.8%
  59. 59 Gemini 3.5 Flash-Lite 29.3%
  60. 60 Kimi K2.6 27.9%
  61. 61 Gemini 3.5 Flash 27.5%
  62. 62 Tencent Hy3 25.6%
  63. 63 Qwen3.5 397B-A17B 24.9%
  64. 64 Gemini 3 Flash high reasoning effort where available 24%
  65. 65 GPT-5.2 high reasoning effort where available 23%
  66. 66 Nemotron 3 Ultra 22.7%
  67. 67 DeepSeek-V3.2 Thinking; Novita 21.3%
  68. 68 Solar Pro 4 18.7%
  69. 69 Gemini 3 Pro high reasoning effort where available 18.4%
  70. 70 Claude Opus 4.5 high reasoning effort where available 18.4%
  71. 71 GPT-5 high reasoning effort where available 18.3%
  72. 72 Solar Open 2 16.6%
  73. 73 Grok 4 high reasoning effort where available 15.2%
  74. 74 gpt-oss-120b high reasoning effort where available 4.70%
  75. 75 Kimi K2 Thinking high reasoning effort where available 4.00%

Sources: Another board, via Epoch AI (CC BY 4.0), labs’ model cards and official APEX-Agents leaderboard

#ModelScoreSettingDateSource
1 Kimi K3.1 Pro 67.2% Oct 2026 Lab model card
2 Grok 4.6 57.5% Aug 2026 Lab model card
3 Kimi K3 41% Jul 2026 Lab model card
4 Step 5 Preview 37.8% Sep 2026 Lab model card
5 Tencent Hy4 preview 37.1% Aug 2026 Lab model card
6 Seed2.1 Pro 33.8% Jun 2026 Lab model card
7 Gemini 3.1 Pro 33.5% Feb 2026 Lab model card
8 Kimi K2.6 27.9% Apr 2026 Lab model card
9 MiniMax-M3 27.7% Jun 2026 Lab model card
10 Tencent Hy3 25.6% Jul 2026 Lab model card
11 Solar Pro 4 18.7% Aug 2026 Lab model card
12 Solar Open 2 16.6% Jul 2026 Lab model card
#ModelScoreSettingDateSource
1 Gemini 3 Flash 24% high reasoning effort where available Dec 2025 Official leaderboard
2 GPT-5.2 23% high reasoning effort where available Dec 2025 Official leaderboard
3 Gemini 3 Pro 18.4% high reasoning effort where available Nov 2025 Official leaderboard
4 Claude Opus 4.5 18.4% high reasoning effort where available Nov 2025 Official leaderboard
5 GPT-5 18.3% high reasoning effort where available Aug 2025 Official leaderboard
6 Grok 4 15.2% high reasoning effort where available Jul 2025 Official leaderboard
7 gpt-oss-120b 4.70% high reasoning effort where available Aug 2025 Official leaderboard
8 Kimi K2 Thinking 4.00% high reasoning effort where available Nov 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 4 Carbon 84.2% Oct 2026 Collected by Epoch AI, source not named
2 Gemini 4 Argon 82.4% Sep 2026 Collected by Epoch AI, source not named
3 Claude Fable 5.2 80.1% Nov 2026 Collected by Epoch AI, source not named
4 Claude Sonnet 5.5 75.5% max Sep 2026 Collected by Epoch AI, source not named
5 Claude Opus 5.5 73.5% max Sep 2026 Collected by Epoch AI, source not named
6 DeepSeek V4.1 Flash 1017 69.8% max Oct 2026 Collected by Epoch AI, source not named
7 Claude Fable 5.1 68.6% Sep 2026 Collected by Epoch AI, source not named
8 GPT-6.1 Terra 68.3% max Oct 2026 Collected by Epoch AI, source not named
9 Gemini 3.7 Flash 67.8% Aug 2026 Collected by Epoch AI, source not named
10 Kimi K3.1 Pro 67.2% Oct 2026 Lab model card
11 Muse Wonder 66.1% max Oct 2026 Collected by Epoch AI, source not named
12 Claude Opus 5 65.8% max Jul 2026 Collected by Epoch AI, source not named
13 Grok 4.6 65.3% Aug 2026 Collected by Epoch AI, source not named
14 GPT-6 Astra 64.7% unknown thinking Sep 2026 Collected by Epoch AI, source not named
15 Gemini 3.8 Flash 64.3% Sep 2026 Collected by Epoch AI, source not named
16 Muse Ember 64.3% Nov 2026 Collected by Epoch AI, source not named
17 Claude Fable 5 63.6% Jun 2026 Collected by Epoch AI, source not named
18 Qwen 3.8 Max 63.3% Aug 2026 Collected by Epoch AI, source not named
19 GPT-6.1 Sol 60% max Sep 2026 Collected by Epoch AI, source not named
20 Muse Spark 1.3 58.6% Sep 2026 Collected by Epoch AI, source not named
21 GPT-5.6 Terra 58.2% max Jul 2026 Collected by Epoch AI, source not named
22 GLM-5.3 56.6% unknown thinking Aug 2026 Collected by Epoch AI, source not named
23 Grok 4.5 56.2% unknown thinking Jul 2026 Collected by Epoch AI, source not named
24 Claude Haiku 5.5 56.1% max Oct 2026 Collected by Epoch AI, source not named
25 GPT-5.5 55.1% unknown thinking Apr 2026 Collected by Epoch AI, source not named
26 Grok 4.7 54.6% Sep 2026 Collected by Epoch AI, source not named
27 Claude Sonnet 5 54.5% unknown thinking Jun 2026 Collected by Epoch AI, source not named
28 MiMo-V2.6 Pro 54.4% Sep 2026 Collected by Epoch AI, source not named
29 GPT-6 Sol 54.3% max Sep 2026 Collected by Epoch AI, source not named
30 GLM-5.3-Flash 52.8% Aug 2026 Collected by Epoch AI, source not named
31 GPT-5.4 52.4% unknown thinking Mar 2026 Collected by Epoch AI, source not named
32 Mistral Large 4 Preview 51.6% high Oct 2026 Collected by Epoch AI, source not named
33 GPT-5.6 Sol 51.4% pro, max Jul 2026 Collected by Epoch AI, source not named
34 Claude Opus 4.7 49.2% max Apr 2026 Collected by Epoch AI, source not named
35 Claude Opus 4.8 48.9% max May 2026 Collected by Epoch AI, source not named
36 Beam 1.0 48.8% max Oct 2026 Collected by Epoch AI, source not named
37 Kimi K3 47.8% Jul 2026 Collected by Epoch AI, source not named
38 Qwen 3.8 27B 47.5% Aug 2026 Collected by Epoch AI, source not named
39 DeepSeek V4 Pro 0813 47.3% Aug 2026 Collected by Epoch AI, source not named
40 Gemini 3.6 Flash 46.9% Jul 2026 Collected by Epoch AI, source not named
41 Inkling 2 Small 46.8% Nov 2026 Collected by Epoch AI, source not named
42 Claude Opus 4.6 46.3% max Feb 2026 Collected by Epoch AI, source not named
43 Muse Spark 1.2 45.4% Aug 2026 Collected by Epoch AI, source not named
44 GLM-5.2 45.2% Jun 2026 Collected by Epoch AI, source not named
45 GPT-6 Luna 44.3% max Sep 2026 Collected by Epoch AI, source not named
46 Claude Sonnet 4.6 43% high Feb 2026 Collected by Epoch AI, source not named
47 GPT-5.6 Luna 43% unknown thinking Jul 2026 Collected by Epoch AI, source not named
48 GLM-5.1 40.9% Apr 2026 Collected by Epoch AI, source not named
49 DeepSeek V4.1 Flash 39.5% Sep 2026 Collected by Epoch AI, source not named
50 Step 5 Preview 37.8% Sep 2026 Lab model card
51 MiniMax-M3 37.7% Jun 2026 Collected by Epoch AI, source not named
52 Kimi K2.7 Code 37.6% Jun 2026 Collected by Epoch AI, source not named
53 Tencent Hy4 preview 37.1% Aug 2026 Lab model card
54 Gemini 3.1 Pro 35.1% Feb 2026 Collected by Epoch AI, source not named
55 Seed2.1 Pro 33.8% Jun 2026 Lab model card
56 Inkling 33.8% Jul 2026 Collected by Epoch AI, source not named
57 Muse Spark Open 33.4% Oct 2026 Collected by Epoch AI, source not named
58 Muse Spark 1.1 31.8% Jul 2026 Collected by Epoch AI, source not named
59 Gemini 3.5 Flash-Lite 29.3% Jul 2026 Collected by Epoch AI, source not named
60 Kimi K2.6 27.9% Apr 2026 Lab model card
61 Gemini 3.5 Flash 27.5% May 2026 Collected by Epoch AI, source not named
62 Tencent Hy3 25.6% Jul 2026 Lab model card
63 Qwen3.5 397B-A17B 24.9% Feb 2026 Collected by Epoch AI, source not named
64 Gemini 3 Flash 24% high reasoning effort where available Dec 2025 Official leaderboard
65 GPT-5.2 23% high reasoning effort where available Dec 2025 Official leaderboard
66 Nemotron 3 Ultra 22.7% Jun 2026 Collected by Epoch AI, source not named
67 DeepSeek-V3.2 21.3% Thinking; Novita Dec 2025 Collected by Epoch AI, source not named
68 Solar Pro 4 18.7% Aug 2026 Lab model card
69 Gemini 3 Pro 18.4% high reasoning effort where available Nov 2025 Official leaderboard
70 Claude Opus 4.5 18.4% high reasoning effort where available Nov 2025 Official leaderboard
71 GPT-5 18.3% high reasoning effort where available Aug 2025 Official leaderboard
72 Solar Open 2 16.6% Jul 2026 Lab model card
73 Grok 4 15.2% high reasoning effort where available Jul 2025 Official leaderboard
74 gpt-oss-120b 4.70% high reasoning effort where available Aug 2025 Official leaderboard
75 Kimi K2 Thinking 4.00% high reasoning effort where available Nov 2025 Official leaderboard

The benchmark's own leaderboard has 8 models. Epoch AI collected claims for 61 models from labs’ reports and papers. Switch the scores above to rank them together.

5 models have held number one in labs’ own numbers since its release in January 2026 Number one over time 12 models have held number one with claims included since its release in January 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.

Google DeepMind ByteDance Seed Moonshot AI xAI
Own leaderboard Not tested since Jan 2026
Gemini 3 Flash 24%
Lab claims 5 leaders
Kimi K3.1 Pro 67.2%

Sources: Official APEX-Agents 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 A claim Epoch AI collected from a report or paper Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon Claude Sonnet 5.5 Claude Fable 5.1 Claude Fable 5 GPT-5.5 Claude Opus 4.6 Gemini 3 Flash 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Gemini 4 Carbon Claude Sonnet 5.5

Sources: Official APEX-Agents leaderboard, labs’ model cards and claims collected by Epoch AI (CC BY 4.0)

6 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
Moonshot AI
MiniMax
Tencent
Upstage
xAI
StepFun
DateLaunch postScoreAs the post writes it
Oct 24, 2026 Kimi K3.1 Pro 67.2 APEX-Agents
Sep 20, 2026 Step 5 Preview 37.8 Apex-Agents
Aug 28, 2026 Tencent Hy4 preview 37.1 APEX-Agents (pass@1)
Aug 12, 2026 Grok 4.6 57.5 APEX-Agents
Aug 11, 2026 Solar Pro 4 18.7 APEX-Agents* (in-house)
Jul 22, 2026 Solar Open 2 16.6 APEX-Agents
Jul 16, 2026 Kimi K3 41 APEX-Agents
Jul 2, 2026 Tencent Hy3 25.6 Apex-Agent (pass@1)
Jun 1, 2026 MiniMax-M3 27.7 Apex-Agents
Apr 20, 2026 Kimi K2.6 27.9 APEX-Agents
Feb 19, 2026 Gemini 3.1 Pro 33.5 APEX-Agents

5 system cards report it, from 5 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
Aug 12, 2026 Grok 4.6 model card 57.5 APEX-Agents, Pass@1, high, Mercor harness (3.3) Capabilities
Jul 27, 2026 Kimi K3 technical report 41 APEX-Agents, cited from Artificial Analysis (Table 2) Capabilities
Jul 22, 2026 Solar Open 2 technical report 16.6 APEX-Agents (Table 3) Capabilities
Jun 23, 2026 Seed2.1 Pro model card 33.8 APEX Agents (Table 1) Capabilities
Feb 19, 2026 Gemini 3.1 Pro model card 33.5 APEX-Agents (results table, p.4) Capabilities

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.

Google DeepMind ByteDance Seed Moonshot AI xAI Anthropic
APEX-Agents 1.1 Passed 85% in Sep 2026, by Gemini 4 Argon
Gemini 4 Carbon 90.4%
APEX-Agents 5 leaders
Kimi K3.1 Pro 67.2%
APEX-Agents 1.1 Passed 85% in Sep 2026, by Gemini 4 Argon
Gemini 4 Carbon 90.4%

Sources: Official leaderboards and labs’ model cards

Replaced by APEX-Agents 1.1

More in Economic work

All 97 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). Claude Fable 5 75.9%
  • 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%