By vendor
OpenAI family
GPT-5.5 / 5.4, o4-mini, o3, GPT-4o, gpt-image-2, gpt-realtime, Whisper, TTS
Anthropic Claude
Claude Opus 4.8, Sonnet 4.6, Haiku 4.5, with thinking / adaptive-thinking modes
Google Gemini
Gemini 3.1 Pro, 3.5 Flash, 3.1 Flash-Lite, Nano Banana, Veo
Image / music
Midjourney (Imagine / Upscale / Blend), Suno, Udio
Embeddings & rerank
text-embedding-3-*, bge, Cohere, Jina Rerank
Chinese vendors
DeepSeek-V3 / R1, Qwen3, GLM-4.6, Doubao, Kimi, Yi and more
By use case
| Use case | Recommended (cheap) | Recommended (high quality) |
|---|---|---|
| General chat | gpt-5.4-mini, claude-haiku-4-5, gemini-3.1-flash-lite | gpt-5.5, claude-opus-4-8, gemini-3.1-pro-preview, gemini-3.5-flash |
| Long context | gemini-3.1-flash-lite (1M) | gemini-3.1-pro-preview (1M), claude-opus-4-8 (1M), claude-sonnet-4-6 (1M), gpt-5.5 (1M) |
| Complex reasoning | o4-mini, deepseek-r1 | gpt-5.5 (high), claude-opus-4-8, o3, gemini-3.1-pro-preview |
| Vision (image understanding) | gpt-5.4-mini, gemini-3.1-flash-lite | gpt-5.5, claude-opus-4-8, gemini-3.1-pro-preview |
| Code | claude-haiku-4-5, gpt-5.4-mini | claude-opus-4-8, gpt-5.5, gemini-3.1-pro-preview |
| Translation | gpt-5.4-mini | claude-opus-4-8 |
| Function calling | gpt-5.4-mini | gpt-5.5, claude-opus-4-8, gemini-3.1-pro-preview |
| Image generation | dall-e-3, imagen-4 | gpt-image-2, mj_imagine, nano-banana-pro |
| Text embeddings | text-embedding-3-small, gemini-embedding-2 | text-embedding-3-large |
| Audio transcription | whisper-1, gpt-4o-mini-transcribe | gpt-4o-transcribe |
| Text to speech | tts-1 | tts-1-hd, gpt-4o-mini-tts, gemini-3.1-flash-tts-preview |
| Realtime voice | — | gpt-realtime-2, gemini-3.1-flash-live-preview |
| Video generation | — | sora-2, veo-3.1 |
Model ID naming conventions
- Official original ID:
gpt-5.5,claude-opus-4-8,gemini-3.1-pro-preview— consistent with each vendor. - Reasoning-effort suffix:
-high/-medium/-lowcontrols the reasoning budget of o3, gpt-5.x, Gemini Thinking, etc. - Thinking toggle:
claude-3-7-sonnet-20250219-thinking,gemini-2.5-flash-thinkingetc. explicitly enable thinking mode. - No-thinking variant:
gemini-2.5-flash-nothinkingetc. explicitly disable thinking. - Image / music specialized:
mj_imagine,mj_upscale,suno_v3.5.
Capability matrix
| Model family | Streaming | Tool calls | Vision | Thinking | JSON mode | Prompt caching |
|---|---|---|---|---|---|---|
| GPT-5.x family | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| GPT-4o family | ✅ | ✅ | ✅ | — | ✅ | ✅ |
| o-series reasoning | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Claude 4.x | ✅ | ✅ | ✅ | ✅ | ✅ (JSON) | ✅ |
| Gemini 3.x / 2.5 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| DeepSeek-V3 / R1 | ✅ | ✅ | — | R1 ✅ | ✅ | — |
| Qwen3 | ✅ | ✅ | VL ✅ | — | ✅ | — |
Model changes & retirement
- Additions: new models are added continuously; the console model list is authoritative.
- Retirement: when an upstream deprecates a model, OpenPAI announces it 7-14 days in advance and stops accepting new requests.
- Price changes: multiplier changes are announced 24 hours in advance; orders already generated are settled at the multiplier in effect at call time.
Next steps
Billing
Charging rules for tokens, multipliers, cache and chain of thought.
API reference
Request / response fields for every endpoint.