Benchmarks · Agents & computer use · Professional work
Company Agent Bench
Does a workplace agent know its company? 205 questions from a month of a real company's Slack: who wrote a message, who was tagged, and who wrote each reply. API models with a simple Slack harness; share answered correctly, with cost per question. From Phil Chen and Josh Qin.
6 models · 18 results · data as of 2026-11-12 · companyagentbench.com ↗
Leaderboard
Fable 5.2 high
Heavy line: best score at each price. Thin lines join one model's effort levels. Cost uses each model's median subscription plan, or API price with no plan data.
All results
18 rows| # | Model | Effort | Company Agent Bench | Subscription $ / task | API $ / task |
|---|---|---|---|---|---|
| 1 | Fable 5.2 | high | 82.2% | $0.049 | $1.34 |
| 2 | GPT-6 Astra | medium | 81.1% | $0.014 | $0.32 |
| 3 | GPT-6 Astra | high | 80.8% | $0.024 | $0.56 |
| 4 | Fable 5.2 | medium | 80.0% | $0.015 | $0.41 |
| 5 | Opus 5.5 | high | 79.2% | $0.007 | $0.44 |
| 6 | GPT-6.1 Sol | high | 79.1% | $0.004 | $0.11 |
| 7 | Fable 5.2 | low | 78.9% | $0.011 | $0.29 |
| 8 | GPT-6.1 Sol | medium | 78.8% | $0.002 | $0.062 |
| 9 | GPT-6.1 Terra | medium | 76.8% | $0.001 | $0.026 |
| 10 | GPT-6 Astra | low | 76.7% | $0.009 | $0.20 |
| 11 | GPT-6.1 Terra | high | 74.3% | $0.001 | $0.036 |
| 12 | GPT-6.1 Sol | low | 73.9% | $0.001 | $0.043 |
| 13 | Opus 5.5 | medium | 72.5% | $0.004 | $0.27 |
| 14 | GPT-6.1 Terra | low | 70.3% | $0.001 | $0.017 |
| 15 | Sonnet 5.5 | high | 68.7% | $0.002 | $0.15 |
| 16 | Opus 5.5 | low | 54.5% | $0.001 | $0.059 |
| 17 | Sonnet 5.5 | medium | 52.7% | $0.001 | $0.048 |
| 18 | Sonnet 5.5 | low | 35.6% | $0.000 | $0.021 |