GPT-6.2 Astra early benchmarks: 78% Terminal-Bench Science, ~$7 per task vs $18-24 for rivals
VraserX · x · 2026-11-18
VraserX rounds up early GPT-6.2 Astra numbers: 78% on Terminal-Bench Science vs 71% for Claude Opus 5.5, 15% vs 4% on Harvey-LAB-AA, and roughly $7 cost per task versus $18-24 for competing models. He also cites his earlier scoop that OpenAI would ship GPT-6.2 Sol and Astra the same day, with GPT-6.5 Bel following two weeks later.
Related event: GPT-6.2 Astra Sets New Record on Terminal-Bench Science(2 posts)→
More from Models
- Microsoft's MAI Models Go Live on Copilot, OpenRouter and Vercel — marlene_zw · 2026-11-18
- Theo's Long Critique: GPT-6.2 Astra Is Painful to Use, Opus 5.5 Still Better Overall — Angaisb_ · 2026-11-18
- GPT-6.2 Astra Is Now Out — nevrekaraishwa2 · 2026-11-18
- Researcher: guidance is underrated and works for weak models — in any space — churchkey · 2026-11-18
- Every Tests OpenAI's New Astra and Sol Models on Coding, Writing and Creative Work — danshipper · 2026-11-18
- Ethan Mollick: GPT-6.2 Is OpenAI's First Frontier Release Since GPT-4.5 That Isn't a Clear Step Forward — emollick · 2026-11-18