Does Headroom currently support Codex model 5.6 (sol, terra, luna)?

[nick1714]

I'm currently using the VS Code extension and Headroom (v0.27.0) via proxy, but when I look at the Headroom dashboard, I see that Headroom isn't cutting any tokens at all. I wonder if the newer version (v0.31.0) supports this? Or is there a configuration error somewhere?

[hosanxiv]

Yes. Current Headroom supports the gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna model names on the Codex Responses/WebSocket path.

The zero shown by v0.27.0 is not sufficient evidence that compression was disabled. Codex uses a long-lived /v1/responses WebSocket, and before v0.29.0 its live compression counters were not included in the main stats summary. Compression could therefore be occurring while the dashboard appeared unchanged.

Please upgrade to v0.29.0 or later, restart the proxy, and inspect summary.codex_ws in /stats:

{
  "units_total": 12,
  "units_modified": 9,
  "tokens_saved": 4321
}

If units_total advances but tokens_saved remains zero, the requests are reaching the Codex compression path but contain no currently eligible savings. If units_total does not advance, that points to a routing/configuration issue instead.

Evidence: PR #1680 documents that Codex WebSocket units were compressed while v0.27-era headline counters could remain frozen. The visibility fix shipped in v0.29.0.