Enterprise interface interaction on WorkArena L2 full benchmark
69.4Success RateGPT-5
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-5Harness=GenericAgent, Params=-2026.04 | 69.4 | |
| Claude 4 SonnetHarness=GenericAgent, Params=-2026.04 | 40.4 | |
| GPT-oss-120BHarness=GenericAgent, Params=120B2026.04 | 11.5 |