Benchmark Almanac

FORTE

GPT-5.5 77.8% leads since Jun 2026 In use

Agents complete 180 office tasks across 15 corporate professions inside a workspace of xlsx, docx, pptx, pdf and other files. An LLM judge checks the output files against expert rubrics, and a task passes only if every rubric item passes.

The repository ships one demo task per profession. The full 180 tasks are run by the maker. Metrics are Avg@3, Pass@3 and Pass^3 with the OpenClaw harness.

Made by AGI-Eval (Meituan) Released Jun 2026 Documents and spreadsheets Public tasks plus a private test set Last in a launch post Jul 2026 Last model added Jul 2026 Scores refreshed 2 Nov 2026
Level
University Undergraduates and entry-level professionals
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.
AGI-Eval-Official/FORTE 20 stars Leaderboard agi-eval-official.github.io
21
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab

GPT-5.5 leads with 77.8% among models released by Jul 2026

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.

The newest model with a score here came out in Jul 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

21 models. Scroll sideways for the rest.

  1. 77.8 GPT-5.5
  2. 77.6 Claude Opus 4.7
  3. 77.2 Claude Opus 4.8
  4. 76.9 GLM-5.2
  5. 73.8 GLM-5.1
  6. 73.2 Claude Opus 4.6
  7. 73.2 LongCat-2.0
  8. 71.2 MiniMax-M3
  9. 70.7 DeepSeek-V4-Flash
  10. 70.5 DeepSeek-V4-Pro
  11. 70.3 Gemini 3.1 Pro
  12. 68.8 GPT-5.4
  13. 68.3 Claude Sonnet 4.6
  14. 63.9 Kimi K2.6
  15. 62.5 Qwen 3.6 Plus
  16. 61.2 GLM-5-Turbo
  17. 60.7 GLM-5
  18. 57.7 Seed2.0 Pro
  19. 55.6 Kimi K2.5
  20. 49.7 Tencent Hy3 preview
  21. 49.5 MiniMax-M2.7
  1. 1 GPT-5.5 77.8%
  2. 2 Claude Opus 4.7 77.6%
  3. 3 Claude Opus 4.8 77.2%
  4. 4 GLM-5.2 76.9%
  5. 5 GLM-5.1 73.8%
  6. 6 Claude Opus 4.6 73.2%
  7. 7 LongCat-2.0 73.2%
  8. 8 MiniMax-M3 71.2%
  9. 9 DeepSeek-V4-Flash 70.7%
  10. 10 DeepSeek-V4-Pro 70.5%
  11. 11 Gemini 3.1 Pro 70.3%
  12. 12 GPT-5.4 68.8%
  13. 13 Claude Sonnet 4.6 68.3%
  14. 14 Kimi K2.6 63.9%
  15. 15 Qwen 3.6 Plus 62.5%
  16. 16 GLM-5-Turbo 61.2%
  17. 17 GLM-5 60.7%
  18. 18 Seed2.0 Pro 57.7%
  19. 19 Kimi K2.5 55.6%
  20. 20 Tencent Hy3 preview 49.7%
  21. 21 MiniMax-M2.7 49.5%
#ModelScoreSettingDateSource
1 GPT-5.5 77.8% OpenClaw Apr 2026 Official leaderboard
2 Claude Opus 4.7 77.6% OpenClaw Apr 2026 Official leaderboard
3 Claude Opus 4.8 77.2% OpenClaw May 2026 Official leaderboard
4 GLM-5.2 76.9% OpenClaw Jun 2026 Official leaderboard
5 GLM-5.1 73.8% OpenClaw Apr 2026 Official leaderboard
6 Claude Opus 4.6 73.2% OpenClaw Feb 2026 Official leaderboard
7 LongCat-2.0 73.2% OpenClaw Jul 2026 Official leaderboard
8 MiniMax-M3 71.2% OpenClaw Jun 2026 Official leaderboard
9 DeepSeek-V4-Flash 70.7% OpenClaw Apr 2026 Official leaderboard
10 DeepSeek-V4-Pro 70.5% OpenClaw Apr 2026 Official leaderboard
11 Gemini 3.1 Pro 70.3% OpenClaw Feb 2026 Official leaderboard
12 GPT-5.4 68.8% OpenClaw Mar 2026 Official leaderboard
13 Claude Sonnet 4.6 68.3% OpenClaw Feb 2026 Official leaderboard
14 Kimi K2.6 63.9% OpenClaw Apr 2026 Official leaderboard
15 Qwen 3.6 Plus 62.5% OpenClaw Mar 2026 Official leaderboard
16 GLM-5-Turbo 61.2% OpenClaw Mar 2026 Official leaderboard
17 GLM-5 60.7% OpenClaw Feb 2026 Official leaderboard
18 Seed2.0 Pro 57.7% OpenClaw Feb 2026 Official leaderboard
19 Kimi K2.5 55.6% OpenClaw Jan 2026 Official leaderboard
20 Tencent Hy3 preview 49.7% OpenClaw Apr 2026 Official leaderboard
21 MiniMax-M2.7 49.5% OpenClaw Mar 2026 Official leaderboard

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

Number one over time

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

OpenAI
Own leaderboard Not tested since Jul 2026
GPT-5.5 77.8%

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% 2026 Near the ceiling (85%) GPT-5.5 Claude Opus 4.6 Kimi K2.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Kimi K2.5

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.

Meituan
DateLaunch postScoreAs the post writes it
5 Jul 2026 LongCat-2.0 73.2 FORTE

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%