Kimi K2.7 RL recap: gains transfer to unseen benchmarks while steps drop ~35%
echen · x · 2026-09-12
Surge AI's detailed recap of RL post-training on Kimi K2.7 (Max reasoning) reports three key findings:
- Gains transfer: training tasks predate DeepSWE, SWE-Marathon, and Terminal-Bench 3; improvements hold across three different agent harnesses with different tools, prompting, and loop structures. SWE-Marathon requires multi-hour system-building trajectories; TB3 spans hardware, scientific computing, systems, and security.
- Smaller model beats larger ones: per K3's self-reported scores, the post-trained K2.7 leads K3 on Terminal-Bench 2.1 and 3, and slightly edges GPT-5.6 Sol on SWE-Bench Pro (contextual, not controlled comparisons).
- More efficient: median trajectory length fell from 150 to 98 steps on DeepSWE and 102 to 78 on Terminal-Bench 3.
Core conclusion: K2.7 already knew how to code—RL made its execution less brittle, turning intern-style code into shippable code. The team studied trajectories to identify the specific new behaviors learned.
Related event: RL Post-Training on 1,700 Tasks Boosts Kimi K2.7 Across Coding Benchmarks(2 posts)→
More from coding & agent
- The legendary fake Go generics: Unicode Canadian Syllabics as angle brackets — rickasaurus · 2026-10-28
- Microsoft Agensh board: a coding agent's "tests passed" is just a message — shoumikgoswami · 2026-10-28
- Ollama and FriendliAI to Host Seoul Meetup on Building Production Agents with Open Models — ollama · 2026-10-28
- theo: If Your Product Requires a Dashboard, Claude Will Just Rebuild It — sull · 2026-10-28
- Magic Patterns' first engineer: Cursor in Slack, a Patterson review agent, repo clones over worktrees — Y Combinator Blog · 2026-10-28
- Dev builds AI job-search app for his wife, she lands a job, and he open-sources it — CurieuxExplorer · 2026-10-28