GPT-6 Astra beats Stockfish limiters at 1320/1500, loses both games at 1700
stayhappyenjoylife · reddit · 2026-09-08
A Redditor ran GPT-6 Astra through six full games vs Stockfish 18 (UCILimitStrength, fixed 200k nodes/move), feeding only the FEN each turn — no engine, legal move list, opening book, tablebase, web, or tools. Result: Astra won both games at 1320 and both at 1500, then lost both at 1700. In one 1700 game it reached +4.22 before blundering. The author cautions limiter settings aren't human Elo and six games prove little; the FEN protocol tests decision-making, not unaided board memory.
A separate preregistered reliability test produced 159/160 legal, well-formed moves. All PGNs, prompts, raw outputs, and a frozen benchmark snapshot are public on GitHub (blunder-zero-ai-chess-benchmark), analyzed with the author's Blunder Zero project.
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26