SWE-Chat v2: largest real-user coding agent dataset grows to 230K prompts from 18K sessions
Diyi_Yang · x · 2026-10-06
SWE-Chat, the largest dataset of real-world coding agent interactions, has released v2 with 230K prompts from 18K sessions. The team is presenting it at COLM this week, noting more traces, new features, and substantial community adoption — v1 has already enabled multiple follow-up research projects.
More from coding & agent
- GXL Announces re:AGENT II, an Agentic Science Hackathon in SF on Jan 9-10 — james_y_zou · 2026-10-06
- DHH's agent-built Campfire benchmarks under fire: Rust was fast because it got hand-tuned — rakyll · 2026-10-06
- SelfBench turns real GitHub PRs into evals: open-weight models cost more and do worse — ycombinator · 2026-10-06
- voice-glow: open-source voice-reactive glow component with built-in emotion recognition — jakubzeg · 2026-10-06
- HashiCorp founder's Superlogical opens public testing for Rex, a free drop-in terminal replacement — jedisct1 · 2026-10-06
- Computer use is the next agent battleground, and Codex Desktop + Astra leads the pack — Yuchenj_UW · 2026-10-06