Auto-derived Codex models and opt-in live tests
- The
openai-codex model list is now built from the OpenAI catalog on every rebuild, so there are no hand-typed entries to keep up to date. This adds gpt-6-sol, gpt-6-luna, gpt-6.1-sol, gpt-daybreak-blue-latest and gpt-daybreak-red-latest.
- Codex pricing now comes from the OpenAI catalog. Every Codex model has a 272k context window, which matches Codex's own default (input above 272k is billed at 2x). You can raise it with
modelOverrides. The build fails with an error if any expected Codex model disappears from the catalog.
- Live provider tests no longer run by default. Set
DREB_LIVE_API=1 to run them. The context-overflow tests also need DREB_LIVE_API_EXPENSIVE=1. DREB_SKIP_LIVE_API has been removed.
- A Copilot
413 Request Entity Too Large response is now treated as a context overflow. The Copilot live tests now pick models that most plans can use.
Implemented in PR 539, resolving issue 538.
