Benchmark Almanac

EnterpriseClawBench

GPT-5.5 66.3% leads since June 2026
Not in model cards

Agents do 852 tasks rebuilt from real proprietary workplace agent sessions, each with fixtures, hard rules and semantic rubrics.

81 HF paper upvotes. The sessions hold internal enterprise content, so not all of it is public.

Made by
Frontis AI
Released
June 2026
Kind
Documents and spreadsheets
Level
Expert
Scored
Graded by a model
Answers
Agent task
Strongest model tested
GPT-5.5 #20 on the capability index
Last model added
June 2026
Scores refreshed
November 12th, 2026

In the authors’ words

Enterprise agents increasingly operate inside workspaces: they read heterogeneous files, invoke tools, and deliver business artifacts. We introduce EnterpriseClawBench, an enterprise agent benchmark constructed from proprietary, real-world agent sessions. Starting from a large archive of workplace sessions, the EnterpriseClawBench produces 852 reproducible tasks, each paired with recovered fixtures, rewritten prompts, role classes, skill subclasses, hard rules, and semantic rubrics. Because the sessions contain internal enterprise content, we do not release the benchmark data; instead, our reusable contribution is the construction and evaluation protocol. On EnterpriseClawBench, the best configuration reaches only 0.663 (Codex with GPT-5.5). These results show that enterprise agent evaluation must report harness–model combinations, artifact delivery, visual quality, cost, runtime, and skill-transfer behavior, rather than collapsing performance into a single score. Code: https://github.com/FrontisAI/EnterpriseClawBench

Abstract of EnterpriseClawBench: Benchmarking Agents from Real Workplace Sessions (arXiv:2606.23654, CC0)
9
Models scored on its own leaderboard

GPT-5.5 leads with 66.3% among models released by June 2026

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.

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

  1. 66.3 GPT-5.5
  2. 64.4 Claude Sonnet 4.6
  3. 63.5 Claude Opus 4.6
  4. 58.8 DeepSeek-V4-Pro
  5. 56.1 Kimi K2.6
  6. 52.5 MiniMax-M3
  7. 51.3 Claude Haiku 4.5
  8. 28.6 GPT-4.1 mini
  9. 27.0 Qwen3-235B-A22B
  1. 1 GPT-5.5 66.3%
  2. 2 Claude Sonnet 4.6 64.4%
  3. 3 Claude Opus 4.6 63.5%
  4. 4 DeepSeek-V4-Pro 58.8%
  5. 5 Kimi K2.6 56.1%
  6. 6 MiniMax-M3 52.5%
  7. 7 Claude Haiku 4.5 51.3%
  8. 8 GPT-4.1 mini 28.6%
  9. 9 Qwen3-235B-A22B 27%

Source: Official EnterpriseClawBench leaderboard

#ModelScoreSettingDateSource
1 GPT-5.5 66.3% Codex Apr 2026 Official leaderboard
2 Claude Sonnet 4.6 64.4% Claude Code Feb 2026 Official leaderboard
3 Claude Opus 4.6 63.5% Claude Code Feb 2026 Official leaderboard
4 DeepSeek-V4-Pro 58.8% OpenClaw Apr 2026 Official leaderboard
5 Kimi K2.6 56.1% Hermes Apr 2026 Official leaderboard
6 MiniMax-M3 52.5% Hermes Jun 2026 Official leaderboard
7 Claude Haiku 4.5 51.3% Claude Code Oct 2025 Official leaderboard
8 GPT-4.1 mini 28.6% Hermes Apr 2025 Official leaderboard
9 Qwen3-235B-A22B 27% Hermes Apr 2025 Official leaderboard

Score vs Runtime

Each point is one model-harness configuration: the primary score against the median time per task. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 0100200300400500 Median time (s), lower is better Primary score GPT-4.1 mini: Primary score 26%, median time 66 s GPT-4.1 mini: Primary score 25.6%, median time 56 s Qwen3-235B-A22B: Primary score 27%, median time 162 s Qwen3-235B-A22B: Primary score 24.1%, median time 137 s Qwen3-235B-A22B: Primary score 20%, median time 120 s Claude Haiku 4.5: Primary score 50.7%, median time 160 s Claude Haiku 4.5: Primary score 42.9%, median time 148 s Claude Haiku 4.5: Primary score 40.5%, median time 145 s Claude Opus 4.6: Primary score 63.5%, median time 215 s Claude Opus 4.6: Primary score 60.2%, median time 238 s Claude Opus 4.6: Primary score 57.3%, median time 283 s Claude Opus 4.6: Primary score 42.5%, median time 272 s Claude Sonnet 4.6: Primary score 64.4%, median time 162 s Claude Sonnet 4.6: Primary score 63.2%, median time 220 s Claude Sonnet 4.6: Primary score 62.3%, median time 210 s Claude Sonnet 4.6: Primary score 45.8%, median time 342 s Kimi K2.6: Primary score 56.1%, median time 462 s Kimi K2.6: Primary score 51.9%, median time 290 s Kimi K2.6: Primary score 51.5%, median time 306 s GPT-5.5: Primary score 62.9%, median time 138 s GPT-5.5: Primary score 62%, median time 167 s GPT-5.5: Primary score 61.4%, median time 137 s DeepSeek-V4-Pro: Primary score 58.8%, median time 315 s DeepSeek-V4-Pro: Primary score 58.5%, median time 343 s DeepSeek-V4-Pro: Primary score 56.2%, median time 264 s MiniMax-M3: Primary score 52.5%, median time 477 s MiniMax-M3: Primary score 52%, median time 353 s MiniMax-M3: Primary score 42.1%, median time 467 s GPT-4.1 mini: Primary score 28.6%, median time 66 s GPT-4.1 mini: Primary score 26.2%, median time 46 s Claude Haiku 4.5: Primary score 51.3%, median time 83 s GPT-5.5: Primary score 66.3%, median time 117 s GPT-5.5Claude Haiku 4.5GPT-4.1 miniGPT-4.1 mini 0% 20% 40% 60% 80% 0100200300400500 Median time (s), lower is better Primary score GPT-4.1 mini: Primary score 26%, median time 66 s GPT-4.1 mini: Primary score 25.6%, median time 56 s Qwen3-235B-A22B: Primary score 27%, median time 162 s Qwen3-235B-A22B: Primary score 24.1%, median time 137 s Qwen3-235B-A22B: Primary score 20%, median time 120 s Claude Haiku 4.5: Primary score 50.7%, median time 160 s Claude Haiku 4.5: Primary score 42.9%, median time 148 s Claude Haiku 4.5: Primary score 40.5%, median time 145 s Claude Opus 4.6: Primary score 63.5%, median time 215 s Claude Opus 4.6: Primary score 60.2%, median time 238 s Claude Opus 4.6: Primary score 57.3%, median time 283 s Claude Opus 4.6: Primary score 42.5%, median time 272 s Claude Sonnet 4.6: Primary score 64.4%, median time 162 s Claude Sonnet 4.6: Primary score 63.2%, median time 220 s Claude Sonnet 4.6: Primary score 62.3%, median time 210 s Claude Sonnet 4.6: Primary score 45.8%, median time 342 s Kimi K2.6: Primary score 56.1%, median time 462 s Kimi K2.6: Primary score 51.9%, median time 290 s Kimi K2.6: Primary score 51.5%, median time 306 s GPT-5.5: Primary score 62.9%, median time 138 s GPT-5.5: Primary score 62%, median time 167 s GPT-5.5: Primary score 61.4%, median time 137 s DeepSeek-V4-Pro: Primary score 58.8%, median time 315 s DeepSeek-V4-Pro: Primary score 58.5%, median time 343 s DeepSeek-V4-Pro: Primary score 56.2%, median time 264 s MiniMax-M3: Primary score 52.5%, median time 477 s MiniMax-M3: Primary score 52%, median time 353 s MiniMax-M3: Primary score 42.1%, median time 467 s GPT-4.1 mini: Primary score 28.6%, median time 66 s GPT-4.1 mini: Primary score 26.2%, median time 46 s Claude Haiku 4.5: Primary score 51.3%, median time 83 s GPT-5.5: Primary score 66.3%, median time 117 s GPT-5.5Claude Haiku 4.5GPT-4.1 miniGPT-4.1 mini
ModelSettingPrimary scoreMedian time
GPT-5.5 Codex 66.3% 117 s
Claude Sonnet 4.6 Claude Code 64.4% 162 s
Claude Opus 4.6 Claude Code 63.5% 215 s
Claude Sonnet 4.6 DeepAgents 63.2% 220 s
GPT-5.5 DeepAgents 62.9% 138 s
Claude Sonnet 4.6 OpenClaw 62.3% 210 s
GPT-5.5 Hermes 62% 167 s
GPT-5.5 OpenClaw 61.4% 137 s
Claude Opus 4.6 OpenClaw 60.2% 238 s
DeepSeek-V4-Pro OpenClaw 58.8% 315 s
DeepSeek-V4-Pro Hermes 58.6% 343 s
Claude Opus 4.6 DeepAgents 57.3% 283 s
DeepSeek-V4-Pro DeepAgents 56.2% 264 s
Kimi K2.6 Hermes 56.1% 462 s
MiniMax-M3 Hermes 52.5% 477 s
MiniMax-M3 DeepAgents 52% 353 s
Kimi K2.6 OpenClaw 51.9% 290 s
Kimi K2.6 DeepAgents 51.5% 306 s
Claude Haiku 4.5 Claude Code 51.3% 83 s
Claude Haiku 4.5 DeepAgents 50.7% 160 s
Claude Sonnet 4.6 Hermes 45.8% 342 s
Claude Haiku 4.5 OpenClaw 42.9% 148 s
Claude Opus 4.6 Hermes 42.5% 272 s
MiniMax-M3 OpenClaw 42.1% 467 s
Claude Haiku 4.5 Hermes 40.5% 145 s
GPT-4.1 mini Hermes 28.6% 66 s
Qwen3-235B-A22B Hermes 27% 162 s
GPT-4.1 mini Codex 26.2% 46 s
GPT-4.1 mini DeepAgents 26% 66 s
GPT-4.1 mini OpenClaw 25.6% 56 s
Qwen3-235B-A22B DeepAgents 24.1% 137 s
Qwen3-235B-A22B OpenClaw 20% 120 s

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 Jun 2026
GPT-5.5 66.3%

Source: Official EnterpriseClawBench leaderboard

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% 2025 2026 Near the ceiling (85%) GPT-5.5 Claude Haiku 4.5 GPT-4.1 mini 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.5 Claude Haiku 4.5

Source: Official EnterpriseClawBench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

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%