Codex vs Claude Code: 30 days of OpenAI releases against Opus 5.5
This week Codex ships many new features, and Claude Code does too. Rafa and Pablo test the two agents. They give a warning about bias: Claude Code is their daily tool for a long time. Javi prefers Codex, but he is not in this stream.
Rafa gives a short history. In 2025, Cursor leads with its agent mode. Claude Code arrives around May and wins the developers who work in the terminal. With Sonnet 4.5 and Opus 4.6, Claude Code takes the lead. Then Opus gets worse, and Codex gets users because you almost never hit its limits. With Opus 5.5, Claude Code uses far fewer tokens, and the situation changes again.
OpenAI replies with 30 days of releases in a row, as Tibo announces. GPT-6 Astra and GPT-6.1 Sol are 50% faster. The Codex subscription works in third-party tools, which Anthropic does not allow. The auto review mode is now free. The Decisions API, remote SSH sessions with Tailscale and instant steering also arrive.
The verdict: the Codex desktop app has a better UX than the Claude Code app. It has good animations, an integrated browser and Image Gen to generate images. But for coding, Opus 5.5 gives better results than any current model, Astra Max included. Pablo misses the Codex steering: Claude Code now sends a message at once with Ctrl+Enter, but you cannot choose the queue as the default. Watch in the video.

All the news we covered