AI model Astra wins 40 of 43 AtCoder contests; Opus 5 dip hints at training contamination
teortaxesTex · x · 2026-09-08
According to wata, problem coordinator for heuristic contests on AtCoder, Astra would place 1st in 40 out of 43 short (4-hour) contests he tested. Another observation: Opus 5 shows a clear performance drop on recent contests, suggesting it may have been explicitly trained on those problems (contamination), while Astra's situation is unclear. It's also unknown whether web search was enabled during tests — if so, models could access solution repos on GitHub, making the results harder to interpret.
More from coding & agent
- Dev calls on OpenAI and Anthropic to donate compute to rewrite C libraries in Rust — cramforce · 2026-09-08
- Commit agent memory to GitHub: markdown files for cross-machine agent continuity — ariG23498 · 2026-09-08
- Lerna plugin routes GitHub Copilot HydraFusion model calls to Azure AI Foundry — unixterminal · 2026-09-08
- Spotify details internal Claude Code setup that cut token usage by 90% — SumitGup · 2026-09-08
- Resurrect dead coding agents: reload the transcript and interrogate the model on why it failed — JoshuaJosephson · 2026-09-08
- AI-built RuneScape clone logs 3,000 signups and 1,000+ play hours in first day — Dimillian · 2026-09-08