catalog
Every model, one price list.
The multiplier next to each id is what it costs against your quota. A 1× model burns one token per token used; a 2× model burns two. Nothing else changes between providers — same key, same endpoint, same response shape.
| Model | Pricing | Modality | Context | Throughput | Latency | 24h status | ID |
|---|---|---|---|---|---|---|---|
| Abliteration FP81 model | |||||||
ablit-fp8-70b | 0.25× | text | 64K | 220 t/s | 240ms | live | ablit-fp8-70b |
| Claude2 models | |||||||
claude-haiku-4.5 | 1× | text · vision | 200K | 132 t/s | 410ms | live | claude-haiku-4.5 |
claude-sonnet-4.5 | 3× | text · vision | 200K | 78 t/s | 640ms | live | claude-sonnet-4.5 |
| DeepSeek3 models | |||||||
deepseek-coder-v3 | 0.6× | text | 128K | 70 t/s | 810ms | live | deepseek-coder-v3 |
deepseek-v3.2 | 0.5× | text | 128K | 64 t/s | 880ms | live | deepseek-v3.2 |
deepseek-r1 | 1.5× | text | 128K | 42 t/s | 1240ms | live | deepseek-r1 |
| Gemini2 models | |||||||
gemini-2.5-flash | 0.5× | text · vision | 1M | 210 t/s | 260ms | live | gemini-2.5-flash |
gemini-2.5-pro | 2× | text · vision | 1M | 88 t/s | 720ms | live | gemini-2.5-pro |
| GLM (Zhipu)2 models | |||||||
glm-4.5-air | 0.4× | text | 128K | 110 t/s | 480ms | live | glm-4.5-air |
glm-4.6 | 1× | text · vision | 200K | 58 t/s | 900ms | live | glm-4.6 |
| Kimi (Moonshot)2 models | |||||||
kimi-k2-turbo | 1.8× | text | 256K | 148 t/s | 520ms | live | kimi-k2-turbo |
kimi-k2 | 1.2× | text | 256K | 52 t/s | 1010ms | live | kimi-k2 |
| MiMo (Xiaomi)1 model | |||||||
mimo-7b-rl | 0.3× | text | 128K | 190 t/s | 300ms | live | mimo-7b-rl |
| OpenAI1 model | |||||||
gpt-5.1 | 2.5× | text · vision | 400K | 96 t/s | 590ms | live | gpt-5.1 |
| Qwen1 model | |||||||
qwen3-max | 1.4× | text · vision | 256K | 74 t/s | 690ms | live | qwen3-max |
multiplier = tokens of quota spent per token used · latency measured p50 over 24h
Streaming by default
Server-sent events pass through unbuffered, including Anthropic's message deltas.
Usage from the source
Token counts come from the provider's usage block, not an estimate we made up.
Drop-in base URL
Swap the base URL in any OpenAI or Anthropic SDK — no wrapper, no shim, no new types.