HN Explorer

Ask HN: Claude Code with Multiple Models?

6 points by iosifnicolae2 ↗ HN
GPT-5 seems stronger at generating diverse ideas, pros/cons, and perspectives, while Sonnet 4.5 feels better at decision-making and writing code. Has anyone tried running them together in Claude Code? Did it improve results?

3 comments

[ 1.9 ms ] story [ 20.6 ms ] thread
covil
I'm asking claude code to collaborate with codex. You can do it by adding codex as a mcp server. They added `codex mcp-server` command for user to use codex as a mcp server recently. And it did improve the plan and code quality; I ask it to review the plan and generated code. Full command is like this: `claude mcp add codex -- codex mcp-server`
moomoo11
Sonnet 4.5 is objectively worse at writing code unless it’s like make me a color picker. IME at least as someone who is going through millions of tokens daily.

Codex high is kind of a beast, even better than Opus which Anthropic decided to limit for their max uses. https://github.com/anthropics/claude-code/issues/8449