Resolve launches SRE-1, a 27B RL-trained model for root-causing production incidents
tianyin_xu · x · 2026-11-19
Resolve announced SRE-1, its first model for finding the root cause of production incidents. On 49 hard cases from live production, it claims the highest pass rate at every budget level, leading the cost-accuracy frontier.
- Built from an open-weight base via reinforcement learning into a custom 27B model
- Trained by a team of 11 researchers using up to 512 GPUs
SRE community member tianyinxu urged Resolve to evaluate on open, reproducible benchmarks like SREGym, following the example of Cognition, Datadog and others, so results can be independently verified.
More from coding & agent
- Soleio ships two Grok agents: Matcher chases intros, Polo auto-drafts inbox replies — soleio · 2026-11-19
- Serial vibe coder: blind vibe coding is a serious security risk for real projects — kylegawley · 2026-11-19
- Vercel launches Board: a secure message board where agent swarms merge 50%+ of v0 PRs — cramforce · 2026-11-19
- From weeks of manual parametrization to one prompt: peptide MD simulation's AI makeover — IgorCarron · 2026-11-19
- Mercury Books goes MCP-native: do your bookkeeping inside Claude, ChatGPT and Cursor — dunkhippo33 · 2026-11-19
- LLM autoresearch boosts Qwen3.8 decode speed 2.4x on RTX 5060 Ti in 20 hours — bobaburger · 2026-11-19