RL on just 1,700 tasks lifts Kimi K2.7 across five coding benchmarks
echen · x · 2026-09-12
Post-training Kimi K2.7 (Max reasoning) with pure RL on just 1,700 Surge coding tasks improved it across all five external benchmarks: +20.0 SWE-Marathon, +14.6 Terminal-Bench 2.1, +12.4 DeepSWE, +10.7 Terminal-Bench 3, +4.7 SWE-Bench Pro.
The framing: before RL, K2.7 could write code like an intern but failed the last mile—dropping requirements, writing narrow tests, creating regressions. After RL it ships like a staff engineer.
Related event: RL Post-Training on 1,700 Tasks Boosts Kimi K2.7 Across Coding Benchmarks(2 posts)→
More from coding & agent
- One-person companies are about the org chart, not headcount: humans judge, agents execute — ciguleva · 2026-10-28
- OATs: open-source toolkit turns everyday objects into personality-bearing agents — Stefania_druga · 2026-10-28
- Dev: local coding dropped from 100% to 5% in a year as cloud agents take over — bytebot · 2026-10-28
- Open-sourced 501(c)(FREE): agent-built free back office for US nonprofits — gralston · 2026-10-28
- Grok Bot 1.0.0 early build reveals Shared Bots, service accounts and team workflows — XFreeze · 2026-10-28
- Anne Griffin hosts 20-person SF salon on the product decisions behind AI agents — annetgriffin · 2026-10-28