Coding agent index: GPT-6.2 Astra flat, GPT-6.2 Sol drops 3 points vs 6.1

Yuchenj_UW · x · 2026-11-18

Yuchen Jin flags Artificial Analysis data questioning OpenAI's upgrade story: GPT-6.2 Astra is flat with GPT-6 Astra (53 points each), and GPT-6.2 Sol actually loses 3 points relative to GPT-6.1 Sol (46 vs 49) on the Artificial Analysis Coding Agent Index (Codex harness at max effort). 'How is this even possible…'

The attached charts show Claude Code combos topping the index at 56-58, GPT-6.2 models stuck mid-pack, and cost-per-task ranging up to $10-14 for Claude Code + Claude Fabble 5.2 (high)—though GPT-6.2 Astra sits reasonably in the value quadrant. A stark contrast with OpenAI's official 'step up in capability' claim.

Related event: GPT-6.2 Cuts Price 20% but Coding Stalls in Third-Party Benchmarks(4 posts)→

Original post →

More from Models

Models channel →