Benchmarks

Measured, not promised

Run on WebArena with the same models and the same tasks. Benchmarked with GPT-5-mini and Qwen3-30B — no frontier model required.

Metric AlohaJet Chrome DevTools MCP Playwright MCP
Task success rate (WebArena) 88% 51% 11%
Average time per task 39s 73s 84s
Average tokens per task 158k 340k

All numbers are from the WebArena benchmark, pooled across GPT-5-mini and Qwen3-30B runs — 10 repetitions per task. Want us to benchmark AlohaJet on your workloads? Request a demo →