GPT-5.6 Sol Ultra in Codex: What Developers Need to Know

GPT-5.6 Sol Ultra runs parallel subagents in Codex, scoring 91.9% on Terminal-Bench 2.1 — covers config changes, real costs, and the macOS crash bug.

GPT-5.6 Sol Ultra runs parallel subagents in Codex, scoring 91.9% on Terminal-Bench 2.1 — covers config changes, real costs, and the macOS crash bug.
Anthropic's Claude now writes over 80% of its own codebase. Here's what recursive self-improvement actually means for your dev workflow — and the one skill that matters most.