Super-adapter lifts SWE-Bench Pro by 10+ points, cuts tool calls up to 73%
yb2698 · x · 2026-11-20
A hands-on evaluation with the super-adapter shows SWE-Bench Pro (hard 51) rising from 45% to 61% on Hermes Agent and 47% to 67% on OpenCode, with pi steady at 74%. Terminal-Bench 2.1 gains 4-6 points across harnesses; the mean over 6 pairs goes from 55.5% to 63.5%. On efficiency, tasks both solve need 53-73% fewer tool calls with thinking off, and 19-48% fewer with it on.
More from coding & agent
- Pydantic hacks Mypy detection: why type checkers crawl and how to fake your way to speed — charliermarsh · 2026-11-21
- MAI Code 2 hands-on: replacing Claude as orchestrator, tops Terminal-Bench — imdigitalashish · 2026-11-21
- PowerShell creator's Meetup: start AI agents in an empty directory, ditch old cost estimates — dfinke · 2026-11-21
- Time awareness and async input: engineering patterns for personal agents — nikunjhanda · 2026-11-21
- Agent memory needs business ontologies, not smarter LLMs, argues founder — _AustinCalvert_ · 2026-11-21
- MCP Events extension finalized in beta, ready for client and server adoption — dsp_ · 2026-11-21