Enterprise AI in Numbers · Capability and cost
Top score on APEX-Agents, the benchmark of professional services work by AI agents
82.4%
82.4% (Gemini 4 Argon) as of September 30, 2026, according to Epoch AI, Benchmarking hub, up from 75.5% on September 28, 2026. The highest pass rate on APEX-Agents, multi-step tasks drawn from investment banking, management consulting and corporate law, achieved by any evaluated model, with its lab.
What it means for leaders
This is the closest public benchmark to the work of analysts, consultants and lawyers. A rising top score widens the set of professional tasks worth piloting with agents under review.
What would change the reading: Pass rates on the hardest task families, and whether open-weight models close the gap.
The trend
Every reading
| Date | Value | Change | Detail | Source |
|---|---|---|---|---|
| 82.4% | +6.9 points | Gemini 4 Argon | Source | |
| 75.5% | +2.0 points | Claude Sonnet 5.5 | Source | |
| 73.5% | +4.9 points | Claude Opus 5.5 | Source | |
| 68.6% | +0.8 points | Claude Fable 5.1 | Source | |
| 67.8% | +2.0 points | Gemini 3.7 Flash | Source | |
| 65.8% | +2.2 points | Claude Opus 5 | Source | |
| 63.6% | +8.5 points | Claude Fable 5 | Source | |
| 55.1% | +2.7 points | GPT-5.5 | Source | |
| 52.4% | +6.1 points | GPT-5.4 | Source | |
| 46.3% | +25.0 points | Claude Opus 4.6 | Source | |
| 21.3% | +16.9 points | DeepSeek-V3.2 | Source | |
| 4.4% | — | gpt-oss-120b | Source |
Definition and method
- What it measures
- The highest pass rate on APEX-Agents, multi-step tasks drawn from investment banking, management consulting and corporate law, achieved by any evaluated model, with its lab.
- Population
- AI models evaluated on APEX-Agents, as compiled by Epoch AI
- Updated
- When new models are evaluated
- Source
- Epoch AI, Benchmarking hub · CC BY 4.0
Caveats
- Scores come from independent evaluators compiled by Epoch AI, not from the labs, and depend on the settings and tools each run used.
- The leader is the best result so far, dated by the release of the model that set it; a model evaluated late can take the lead after its release date.
Cite this
Enterprise AI in Numbers, "Top score on APEX-Agents, the benchmark of professional services work by AI agents", https://lucianoayres.com/ai-brief/data/professional-work-benchmark-leader/, retrieved October 25, 2026. Data: Epoch AI, Benchmarking hub.