feat(openai): add pricing for the Codex and chat-latest models

[alexander-bal]

What

Adds provider entries for seven OpenAI models that have models/openai/ metadata but no providers/openai/models/ entry, so they never appear under the openai provider in api.json and consumers have no price for them:

model	input	cached input	output
gpt-5-codex	$1.25	$0.125	$10.00
gpt-5.1-codex	$1.25	$0.125	$10.00
gpt-5.1-codex-max	$1.25	$0.125	$10.00
gpt-5.1-codex-mini	$0.25	$0.025	$2.00
gpt-5.2-codex	$1.75	$0.175	$14.00
gpt-5-chat-latest	$1.25	$0.125	$10.00
gpt-5.1-chat-latest	$1.25	$0.125	$10.00

The gpt-5.3 generation onward is already covered — gpt-5.3-codex, gpt-5.3-codex-spark, gpt-5.2-chat-latest and gpt-5.3-chat-latest all have provider entries. These seven are the ones below that line.

Source

Prices are from OpenAI's published price list, standard tier: https://developers.openai.com/api/docs/pricing

Each file cites it in a leading comment block, per the guideline that only a leading header survives the sync rewrite.

Notes

Every entry uses base_model, since models/openai/<id>.toml exists for all seven — no provider-agnostic facts are duplicated inline.

reasoning_options is declared because the inherited metadata sets reasoning = true and the schema requires it. Only the gpt-5.2-codex model page publishes its accepted effort values (low, medium, high, xhigh), so that one is filled in; the other six use reasoning_options = [] per the guideline to leave it empty rather than assert a control that is not documented. Happy to fill those in if you have a source I missed.

bun validate passes.

[github-actions]

Action items

• [medium] [possible mistake] providers/openai/models/gpt-5-codex.toml:8 - Check: reasoning_options must reflect the OpenAI API surface for each reasoning model, using same-model / same-surface peers as review context. Why: These five Codex entries set reasoning_options = [], but no existing providers/openai reasoning model uses an empty list, and same-ID Azure peers (OpenAI-compatible) already declare effort: gpt-5-codex / gpt-5.1-codex / gpt-5.1-codex-mini → low|medium|high; gpt-5.1-codex-max → low|medium|high|xhigh. OpenRouter/Vercel/llmgateway agree. Empty options will tell consumers OpenAI exposes no effort control when peers indicate it does. Action: Verify OpenAI’s accepted reasoning.effort / reasoning_effort values for each model and replace [] with the verified effort list (start from the Azure/OpenRouter peers above; do not invent none/xhigh without evidence). Apply the same fix in gpt-5.1-codex.toml, gpt-5.1-codex-max.toml, and gpt-5.1-codex-mini.toml.
• [low] [possible mistake] providers/openai/models/gpt-5-chat-latest.toml:8 - Check: Chat-latest reasoning controls on the native OpenAI provider. Why: gpt-5-chat-latest and gpt-5.1-chat-latest use reasoning_options = [] while same-provider sibling gpt-5.2-chat-latest declares effort = ["medium"], and no other OpenAI provider reasoning model uses []. Gateway peers also use empty lists, so this may be correct if effort is fixed/undocumented—but it is inconsistent with the chat-latest pattern already on this provider. Action: Confirm against OpenAI model/API docs whether these IDs accept any effort values (including a fixed medium); if they do, declare them; only keep [] if no caller control is actually accepted.

[rekram1-node]

Thanks for the PR — prices look correct against the specialized models table, and the base_model shape is fine. Closing because these models are no longer served on the OpenAI API.

They were intentionally removed in #3707 after OpenAI's 2026-07-23 shutdown (deprecations, 2026-04-22 notice):

Model	Suggested replacement
gpt-5-codex	gpt-5.6-sol
gpt-5.1-codex	gpt-5.6-sol
gpt-5.1-codex-max	gpt-5.6-sol
gpt-5.1-codex-mini	gpt-5.6-terra
gpt-5.2-codex	gpt-5.6-sol
gpt-5-chat-latest	gpt-5.6-sol
gpt-5.1-chat-latest	gpt-5.6-sol

Re-adding them would put dead IDs back in the openai provider catalog. models/openai/ metadata can stay for other gateways that still proxy these IDs; we only drop fully shut-down models from providers/openai.