CaptchaArena releases 50K verified CAPTCHA trajectories; 9B agent hits 71.7% Pass@1
ColumbiaUniversity · hf · 2026-09-30
CaptchaArena is the first large-scale fine-grained training dataset for interactive CAPTCHA solving: 50K puzzles across 20 types and 5 interaction modes, all execution-verified, with 50K screenshot-action trajectories (46K with step-by-step reasoning) and pixel-mask annotations. CaptchaAgent, a single 9B policy trained with SFT plus RL (environment verifier as reward), achieves 70.5% Pass@1 after SFT and 71.7% after RL, also improving on two external benchmarks. Released open source.
More from coding & agent
- Auto-research loop on 120 B300s finds 40% Kimi K3 inference gain for $9,176 — bookwormengr · 2026-09-30
- LENINROOMS: Building an Infinite Soviet Apartment Backrooms With Claude — teortaxesTex · 2026-09-30
- dart-query MCP server adds DartQL bulk task operations to cut token usage — modelcontextprotocol · 2026-09-30
- ISIR MCP connector lets agents query Czech insolvency register by company or person — modelcontextprotocol · 2026-09-30
- New Yorker-style illustration Skill hits 400 stars, monetizes via Baidu agent ecosystem — oran_ge · 2026-09-30
- Dev builds a Grok-powered 'X newsroom' that saves 3 hours a day on posting — jamestagg · 2026-09-30