Web Browsing Automation on WebArena (test)
3.9Latency (s)AgentTrek-32B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| AgentTrek-32BSmall=AgentTrek-32B2026.04 | 3.9 | 0.005 | 30.3 | 16 | — | — | — | |
| gpt-oss-20bSmall=gpt-oss-20b2026.04 | 6.9 | 0.001 | 37.6 | 8.1 | — | — | — | |
| Step-level cascade (gpt-oss-20b + GPT-5 mini)Small=gpt-oss-20b, Large=GPT-5 mini2026.04 | 8.5 | 0.027 | 51.3 | 12.1 | 568 | 30.6 | 69.4 | |
| Step-level cascade (AgentTrek-32B + GPT-5 mini)Small=AgentTrek-32B, Large=GPT-5 mini2026.04 | 9.4 | 0.028 | 52.3 | 8.2 | 562 | 40.1 | 59.9 | |
| GPT-5 miniLarge=GPT-5 mini2026.04 | 11.1 | 0.053 | 55 | 13.8 | — | — | — | |
| Step-level cascade (gpt-oss-20b + GPT-5.2)Small=gpt-oss-20b, Large=GPT-5.22026.04 | 12.2 | 0.211 | 57.8 | 10.3 | 578 | 33.1 | 66.9 | |
| Step-level cascade (AgentTrek-32B + GPT-5.2)Small=AgentTrek-32B, Large=GPT-5.22026.04 | 13.4 | 0.208 | 58.8 | 12 | 593 | 43.7 | 56.3 | |
| GPT-5.2Large=GPT-5.22026.04 | 19.6 | 0.335 | 60.1 | 9.9 | — | — | — |