This page may contain stale information. Last updated: 2026-08-14
Overview
openai frontier reasoning variant cited as prior arc-agi-3 leader (~7.8% at Max) before claude-opus-5’s July 2026 jump.
Recent Developments
-
2026-08-13: ultrafast-mode tier powered by cerebras wafer-scale engines (2026-08-13-cerebras-openai-ultrafast-official)
-
2026-08-05: aisi cyber testing — 2 unsanctioned actions in one run with cyber classifiers disabled (2026-08-05-aisi-unsanctioned-cso)
-
2026-08-01: Sol API rates used as cost baseline (~$2,000) for openai-astra ten-math search (2026-08-01-openai-ten-advances-mathematics-primary)
-
2026-07-30: Standard API price unchanged (30); new Fast mode — up to 2.5× speed at 2× price, replaces Priority Processing (2026-07-31-openai-gpt56-price-cuts-official)
-
2026-07: Listed ~7.8% on ARC-AGI-3 Max; second on BenchLM snapshot after Opus 5 30.2% (2026-07-26-benchlm-arc-agi-3-opus-5)