GPT-6 Astra versus GPT-5.6 Sol on Terminal-Bench 4.0

retained text

Artificial Analysis Coding Agent Index ... In Codex, GPT-6 Astra ... GPT-5.6 Sol ... Terminal-Bench v4.0 (56% vs 37%) ... DeepSWE (68% vs 72%).

document fetched 2026-09-11 10:48