Codex Relay, Third-Party API and Responses API Test Guide
Configure a Codex relay by checking the base URL, model ID, key group and generation endpoint, then review protocol-level paid request samples across six AI relay test accounts.
There is no single Codex relay protocol
A provider may expose the Responses API, Chat Completions, another OpenAI-compatible interface, or a client-specific gateway. Model discovery does not prove the required generation endpoint works.
Configure the actual workflow
Fix the base URL, authentication format, model ID, key group, and generation endpoint before testing streaming, tools, and returned usage fields.
Keep protocol evidence separate
The table retains protocol labels, outcomes, and dates from paid GPT test records. Responses and generic OpenAI samples are not treated as equivalent compatibility evidence.
Codex and Responses API evidence snapshot
- Providers with GPT samples
- 6
- Responses-tested providers
- 3
- Responses successful / total
- 47 / 60
- Latest GPT test
- 2026-09-29
| Provider | Observed protocols | Tested model IDs | Passed / total | Latest test | Published GPT-5.6 Sol group |
|---|---|---|---|---|---|
| UU API | Responses | gpt-6-astra / gpt-5.6-terra / gpt-5.6-sol | 19 / 24 | 2026-09-29 | Codex-GPT Pro account pool |
| H API | Responses | gpt-5.6-terra / gpt-5.6-sol / gpt-6-astra / gpt-5.5 | 7 / 15 | 2026-09-29 | GPT Pro |
| boxying | OpenAI | gpt-6-astra / gpt-5.6-terra / gpt-5.6-sol | 21 / 21 | 2026-09-29 | Codex premium pool |
| LinksAPI | Responses | gpt-6-astra / gpt-5.6-sol / gpt-5.6-terra | 21 / 21 | 2026-09-29 | GPT Pro (Codex included), 0.3x |
| OpenOx | OpenAI | gpt-5.6-sol / gpt-5.6-terra / gpt-6-astra | 21 / 21 | 2026-09-29 | Pro direct route |
| APINebula | OpenAI | gpt-6-astra | 19 / 19 | 2026-09-29 | CODEX |
Protocol names come from the saved test configuration. A Responses-labeled success is endpoint evidence for that model and key, not proof that every Codex client feature works. A generic OpenAI label is not silently upgraded to Responses API support.
Five checks for Codex relay configuration
Base URL
Use the provider's current HTTPS base URL for the exact route group. Do not copy an endpoint from another product line.
Generation endpoint
Confirm whether the route uses /v1/responses, /v1/chat/completions, or a client-specific gateway.
Model ID and key group
Run the exact model required by the project and record the group assigned to the test key.
Streaming and tools
Test streaming, tool calls, and structured output separately when the Codex workflow depends on them.
Usage and billing
Preserve returned token fields and reconcile a controlled request against the account meter or wallet before accepting a published rate.
Evidence boundary
These records verify dated API requests from paid test accounts. They do not establish model identity, every route group, persistent uptime, or the final wallet charge. Use the linked provider record and billing report for those separate checks.
Frequently asked questions
How should I configure a Codex relay?
Confirm the provider's HTTPS base URL, authentication format, exact model ID, route group, and whether the key supports the Responses API or only another OpenAI-compatible endpoint. Start with a low-balance key and one generation request.
Does an OpenAI-compatible API automatically support Codex?
No. Model discovery, Chat Completions, the Responses API, streaming, tools, and the Codex client may use different endpoints or permissions. Test the exact workflow you intend to use.
What does a Responses-labeled sample prove?
It proves that the recorded model, key, route, and Responses test completed at that time. It does not prove every Codex client feature, model identity, long-term availability, or reconciled billing.
Can I compare Codex pools with general GPT API routes?
Only after separating client-only pools, direct routes, reverse routes, protocols, quotas, and billing rules. A low-priced Codex-only pool should not share one price ranking with a general API route.
Answer summary
- Latest test date
- Providers covered
- 6
- Models / protocols
- gpt-5.5 / gpt-5.6-sol / gpt-5.6-terra +1 more · OpenAI / Responses
- Successful / total samples
- 108 / 121
- Price / billing verified
- 6/6 price · 5/6 billing
- Still unverified
- Model identity, every Codex client feature, future uptime and unmatched billing
- One-sentence conclusion
- A GPT or Responses success verifies that dated request only; confirm the exact Codex client, route group and billing rule before buying.