GPT-6 Sol and Luna in Codex: keep your Sume MCP server across models

Switching Codex to GPT-6 Sol or Luna does not change your Sume MCP setup: the server is registered in Codex's own configuration, and every model you pick in the picker sees the same tools. Register https://mcp.sume.com/mcp once with codex mcp add, sign in, and the model choice becomes a separate decision.

Codex's changelog, read 2026-09-29, has entries dated Sep 23 (0.156.1) and Sep 25 (0.157.0) that mention GPT-6 Sol and Luna (0.156.1 says to choose them from the model picker). OpenAI's latest-model guide describes the three GPT-6 models as Astra, Sol and Luna.

How does OpenAI describe the three GPT-6 models?

The guide gives each a one-line role. The table repeats it without ranking them.

How do I register Sume once in Codex?

Codex's MCP page documents codex mcp add <name> --url ... for a streamable HTTP server and codex mcp login for OAuth. Sume's quickstart says to point a streamable HTTP MCP server at the endpoint and run the client's OAuth login for that server name.

codex mcp add sume --url https://mcp.sume.com/mcp
codex mcp login sume

What does OAuth let the model do?

Default OAuth is read-only (mcp:read): discovery, catalog and job reads work. Turn Write on at the consent page to add mcp:write, which the paid tools generate_video and generate_image need. There is no mcp:paid scope.

If you would rather not use a browser sign-in for an unattended run, an API key gives the full tool set; see the separate note on passing it through an environment variable.

Do the tool timeouts change with the model?

No. Codex's MCP page lists tool_timeout_sec with a default of 60 seconds per server. jobs_wait on Sume waits in slices of up to 55 seconds, so it fits inside the default; a longer clip means waiting again, not raising the timeout.