provider
OpenAI
Direct answerOpenAI recommends Responses API for new projects. Direct and gateway calls both use a Bearer header, but require separate OpenAI and site keys, and the gateway endpoint must be implemented for the selected model.
Updated · Reviewed
Official positioning
OpenAI's developer platform spans general text and reasoning, code, vision, image generation, video, speech and realtime interaction, embeddings, and moderation. The official model catalog is the authoritative source for model IDs, modalities, and lifecycle status; resources from the same provider do not automatically share request fields.
Prefer Responses for new projects
OpenAI recommends Responses API for all new projects. Chat Completions remains supported for existing messages and choices integrations. Responses uses input and typed output items, so a migration must update response parsing rather than changing only the URL.
Direct versus gateway authentication
Direct OpenAI calls use the OpenAI base URL, an OpenAI-issued API key, and Authorization: Bearer. This site also recommends a Bearer header, but requires this site's base URL, a site-issued key, and the exact marketplace model ID. The credentials are not interchangeable, and the gateway guarantees only endpoints explicitly listed by its protocol guide and marketplace.
Tools, streaming, and response reading
A custom-function loop requires the application to validate arguments, execute the tool, return its result, and continue the model call. Responses and Chat use different SSE event shapes; parse complete typed events rather than arbitrary network chunks. Whether streaming usage appears or needs an explicit option must be verified per endpoint and channel.
Workloads and production checks
OpenAI models fit structured output, advanced reasoning, code generation, multimodal understanding, tool workflows, and realtime voice. Evaluate quality, latency, and cost on real tasks, then verify snapshots, context, output limits, tools, rate limits, geography, data controls, and deprecation dates before production.
Reviewed provider catalog
Latest models by capability
These are provider-catalog models, not a promise of availability on this site. Confirm callable IDs, endpoints, and pricing in the model marketplace.
Text, reasoning, and code
-
gpt-5.6-solStableThe flagship GPT-5.6 reasoning and coding model for demanding agents, tool use, and long-running tasks.
-
gpt-5.6-terraStableThe balanced GPT-5.6 tier, positioned between maximum capability, latency, and cost.
-
gpt-5.6-lunaStableThe lightweight GPT-5.6 tier for general workloads that prioritize latency and throughput.
Image generation and editing
-
gpt-image-2StableThe latest GPT Image model for text-to-image generation, reference editing, and high-fidelity reference-image input.
Video generation and editing
-
sora-2LegacyThe Sora video model for short video with synchronized audio from text or visual inputs; the official catalog currently marks it as legacy.
-
sora-2-proLegacyThe higher-quality Sora 2 tier for richly detailed video with synchronized audio from text or images; the official catalog marks it as legacy.
Speech and audio
-
gpt-realtime-2.1StableA realtime multimodal model for low-latency spoken conversations, reasoning, and tool use.
-
gpt-realtime-2.1-miniStableThe lightweight GPT Realtime 2.1 tier for realtime voice applications that prioritize cost and throughput.
-
gpt-audio-1.5StableA conversational model for audio input and audio output through Chat Completions.
-
gpt-realtime-translateStableA streaming speech-to-speech translation model for live multilingual audio experiences.
-
gpt-4o-mini-ttsStableA lightweight text-to-speech model for converting text into natural-sounding speech.
-
gpt-transcribeStableA high-accuracy speech-to-text model for file and realtime input transcription.
-
gpt-live-transcribeStableA streaming speech-to-text model that returns low-latency transcript deltas from live audio.
Embeddings, retrieval, and reranking
-
text-embedding-3-largeStableOpenAI's higher-capability text embedding model for semantic search, clustering, and RAG.
-
text-embedding-3-smallStableA cost- and throughput-oriented text embedding model.
Use cases
- Advanced reasoning and coding
- Vision and image generation
- Realtime voice and tool use
API protocols
/v1/chat/completions/v1/responses
FAQ
Should a new OpenAI project use Responses or Chat Completions?
OpenAI recommends Responses for all new projects. Chat Completions remains supported for existing messages/choices integrations. Through this site, also verify that the target model and channel support the endpoint.
Can direct OpenAI and this gateway use the same API key?
No. Both use Authorization: Bearer, but direct calls require an OpenAI-issued key, while this gateway requires a key created by this site and a different base URL.
Does a ChatGPT subscription include OpenAI API access?
Usually not. The chat product and developer API are separate products with separate credentials, balances, and limits. Calling this gateway additionally requires a site API key.
Is Responses text always in output[0]?
No. Responses returns typed output items that can interleave messages, tool calls, and other items. SDK output_text is convenient for simple text, while general clients should dispatch output by type.
Should an OpenAI 429 be retried indefinitely?
No. Honor Retry-After or use bounded exponential backoff for transient limits. Credits, project spend caps, and quota conditions that require user action do not recover through retries.
Related guides
Official sources
- OpenAI Models Official
- OpenAI API Reference Official
- OpenAI Developer Quickstart Official
- Migrate to the Responses API Official
- OpenAI Function Calling Official
- OpenAI Streaming Responses Official
- OpenAI Rate Limits Official
兔子API