Company Agent Bench
205 questions on how well an agent understands its company, built from a month of real Slack history: who wrote an anonymized message, whom a teammate tagged, and who sent each reply. Agents search a replayed sandbox workspace using only messages from before the answer.
Built from one small company's workspace, about 10 teammates. Products ran once on default settings. API models ran in Concept's own Slack harness. The overall score weights the three parts equally.
- Made by
- Concept
- Released
- November 2026
- Kind
- Simulated companies
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- Claude Fable 5.2 #1 on the capability index
- Last model added
- November 2026
- Scores refreshed
- November 12th, 2026
Claude Fable 5.2 leads with 82.2%
Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 1 Claude Fable 5.2 high reasoning effort 82.2%
- 2 GPT-6 Astra medium reasoning effort 81.1%
- 3 Claude Opus 5.5 high reasoning effort 79.2%
- 4 GPT-6.1 Sol high reasoning effort 79.1%
- 5 GPT-6.1 Terra medium reasoning effort 76.8%
- 6 Dots default settings 75.9%
- 7 Grok Bot default settings 72.7%
- 8 Claude Sonnet 5.5 high reasoning effort 68.7%
- 9 Claude Tag default settings 67.8%
- 10 Devin default settings 66.1%
Source: Official Company Agent Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | 82.2% | high reasoning effort, Concept Slack harness | Nov 2026 | Official leaderboard |
| 2 | GPT-6 Astra | 81.1% | medium reasoning effort, Concept Slack harness | Sep 2026 | Official leaderboard |
| 3 | Claude Opus 5.5 | 79.2% | high reasoning effort, Concept Slack harness | Sep 2026 | Official leaderboard |
| 4 | GPT-6.1 Sol | 79.1% | high reasoning effort, Concept Slack harness | Sep 2026 | Official leaderboard |
| 5 | GPT-6.1 Terra | 76.8% | medium reasoning effort, Concept Slack harness | Oct 2026 | Official leaderboard |
| 6 | Dots | 75.9% | default settings | Official leaderboard | |
| 7 | Grok Bot | 72.7% | default settings | Official leaderboard | |
| 8 | Claude Sonnet 5.5 | 68.7% | high reasoning effort, Concept Slack harness | Sep 2026 | Official leaderboard |
| 9 | Claude Tag | 67.8% | default settings | Official leaderboard | |
| 10 | Devin | 66.1% | default settings | Official leaderboard |
The best score, 82.2%, costs $1.34 a task
Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at medium reasoning effort comes within 3.4 points of the best for $0.06 a task, 22 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official Company Agent Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Fable 5.2 | high reasoning effort | Concept Slack harness | 82.2% | $1.34 | |
| GPT-6 Astra | medium reasoning effort | Concept Slack harness | 81.1% | $0.32 | |
| GPT-6 Astra | high reasoning effort | Concept Slack harness | 80.8% | $0.56 | |
| Claude Fable 5.2 | medium reasoning effort | Concept Slack harness | 80% | $0.41 | |
| Claude Opus 5.5 | high reasoning effort | Concept Slack harness | 79.2% | $0.44 | |
| GPT-6.1 Sol | high reasoning effort | Concept Slack harness | 79.1% | $0.11 | |
| Claude Fable 5.2 | low reasoning effort | Concept Slack harness | 78.9% | $0.29 | |
| GPT-6.1 Sol | medium reasoning effort | Concept Slack harness | 78.8% | $0.06 | |
| GPT-6.1 Terra | medium reasoning effort | Concept Slack harness | 76.8% | $0.03 | |
| GPT-6 Astra | low reasoning effort | Concept Slack harness | 76.7% | $0.20 | |
| GPT-6.1 Terra | high reasoning effort | Concept Slack harness | 74.3% | $0.04 | |
| GPT-6.1 Sol | low reasoning effort | Concept Slack harness | 73.9% | $0.04 | |
| Claude Opus 5.5 | medium reasoning effort | Concept Slack harness | 72.5% | $0.27 | |
| GPT-6.1 Terra | low reasoning effort | Concept Slack harness | 70.3% | $0.02 | |
| Claude Sonnet 5.5 | high reasoning effort | Concept Slack harness | 68.7% | $0.15 | |
| Claude Opus 5.5 | low reasoning effort | Concept Slack harness | 54.5% | $0.06 | |
| Claude Sonnet 5.5 | medium reasoning effort | Concept Slack harness | 52.7% | $0.05 | |
| Claude Sonnet 5.5 | low reasoning effort | Concept Slack harness | 35.6% | $0.02 |
2 models have held number one on the official leaderboard since its release in November 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Company Agent Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Company Agent Bench 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
- 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%