Astra saturates Epoch AI's EBR-bench with 100% score in two playthroughs, beating top human
Jsevillamol · x · 2026-09-08
Epoch AI reports that agent Astra has saturated its long-horizon gaming benchmark EBR-bench, scoring 100% with just two playthroughs and outpacing the benchmark's top human player. Both Astra and the top human discovered a card known to 'break the game', and Astra had found a similar cheese strategy in an earlier game release. The benchmark can no longer differentiate agent capabilities.
More from Models
- GPT-6 Astra shows the best instruction-following since Opus 4.5 — 0xSammy · 2026-09-08
- Astra beats GPT-5.6-Sol at personalized digital-avatar shopping, ~4x faster — _AustinCalvert_ · 2026-09-08
- danluu evaluates how well agents use test/verification techniques; Astra impresses — teortaxesTex · 2026-09-08
- Local LLM community wants Gemma 5 to stay chat-first and avoid the Qwen benchmark-chasing trap — AnimalPuzzleheaded71 · 2026-09-08
- Codex users report banked resets failing to arrive on schedule — No-Construction2209 · 2026-09-08
- Anthropic adding usage details button and Claude Code settings to iOS app — testingcatalog · 2026-09-08