ARC-AGI-3 sees a new SOTA at 30.2% as the debate shifts to base models and synthetic data
teortaxesTex · x · 2026-07-26
- A repost of a long discussion around ARC-AGI scores argues that the latest results are more about stronger base models and synthetic-data pipelines than about some magical new reasoning trick.
- It says the human baseline on ARC-AGI-2 was surpassed by the latest generation models, including Gemini 3 Pro and GPT-5.2, and credits Google/OpenAI for most of the gain.
- The quoted ARC Prize update says Claude Opus 5 is the new ARC-AGI-3 SOTA at 30.2%, up from a previous high of 7.8% by GPT-5.6 Sol (Max), and notes novel behaviors that unlocked previously unbeaten environments.
Related event: Claude Opus 5 Sets New ARC-AGI-3 Record(15 posts)→
More from Models
- GPT-6.1 Terra spotted in the OpenAI API, teased by benchmark group LuminaBench — airesearch12 · 2026-10-27
- Ortet's first health eval: 34 open-weight models, 50 tasks, generic rankings fall short — thegautamkamath · 2026-10-27
- Fern 0.1: a 0.1–0.5T coding model one researcher pretrained in 3.5 months for under $10M — teortaxesTex · 2026-10-27
- VLM Run launches OCR Playground for side-by-side comparison of GLM-OCR, DeepSeek-OCR-2 and more — spillai · 2026-10-27
- GPT-6.1-Terra Live With Big Price Cut: $5 vs $12 Output Pricing — scaling01 · 2026-10-27
- Nat Lambert: traces from mid-tier open models are almost all super janky — natolambert · 2026-10-27