an open-market inference gateway
Same models.
Surplus prices.
Sellers dump spare AI capacity on the open market at up to −99%. magpie is one OpenAI-compatible key that routes every request to the cheapest live offer among the models you choose — under the price you set.
Get a key 136 models live · avg discount −78%
The discount window
live| model | in / out per 1M | list | now |
|---|---|---|---|
| MiniMax M2.1MiniMax | $0.0035 / $0.015 | $1.5 | −99% |
| Qwen 3.6 35B A3B (E2EE)Alibaba | $0.0050 / $0.015 | $1.5 | −99% |
| Qwen 3 Next 80bAlibaba | $0.0015 / $0.015 | $1.5 | −99% |
| Mistral Small 3.2 24B InstructMistral AI | $0.0050 / $0.02 | $2 | −99% |
| Trinity Large ThinkingArcee AI | $0.0030 / $0.01 | $1 | −99% |
| Mistral Small 4Mistral AI | $0.0019 / $0.0075 | $0.75 | −99% |
| GPT-5.2 CodexOpenAI | $0.0175 / $0.14 | $14 | −99% |
| Venice MediumMistral AI | $0.0050 / $0.015 | $1.5 | −99% |
prices refresh every 60s from the open marketplace · output $/1M tokens
01
Pick your flock
Choose the models you trust — flagship-only, budget coders, or one key per job. Set a price ceiling, or none.
02
Point your SDK
Swap one base URL. Any OpenAI client works as-is — streaming, chat, completions.
03
We chase the deal
Every request routes to the cheapest healthy offer in your set, live. The response tells you what you paid.
# your existing code, one line different client = OpenAI( base_url="https://magpie.example/v1", api_key="sk-mag-…", ) r = client.chat.completions.create( model="auto", # ← cheapest model in your set, right now messages=[…], )