TheAgentCompany leaderboard
- DeepSeek V3.2 Exp leads
- 42.9%
- models measured
- 4
- lead over #2
- +34.3 pts
- median
- 7.8%
Leader on TheAgentCompany · read October 21
On October 21, 2026, 4 of our catalog's models had a TheAgentCompany result, and DeepSeek V3.2 Exp led with 42.9%.
Work tasks inside a simulated software company: browsing, coding, running programs and messaging colleagues. Share of tasks resolved.
Role in the index
We collect this benchmark and show it as supporting evidence. It is not part of the Vibe Coding Index, because it measures something other than coding and the index is about coding. Read the methodology
| # | Model | Result | Average cost per task | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | DeepSeek V3.2 ExpDeepSeek | 42.9% | $0.40 | 55.0 | $0.41 | |
| 2 | GPT-4o (2024-11-20)OpenAI | 8.6% | $1.29 | 18.2 | $10.00 | |
| 3 | Llama 3.1 70B InstructMeta Llama | 6.9% | $0.93 | Not benchmarked | $0.40 | |
| 4 | Qwen2.5 72B InstructQwen | 5.7% | $1.53 | Not benchmarked | $0.40 |
Only models in our catalog are listed, and each appears once with its best published run. 197 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out. Average cost per task is the average spend on one task, in US dollars, as Epoch AI records it beside the result.