Inference is an OpenAI-compatible gateway for large language models. Point your base_url at us and the same SDK, the same key, reaches OpenAI, Anthropic, Google Gemini and more — without touching a line of your code.
Prepaid credits · no subscription · no contract · revoke keys anytime
curl https://api.alphacurve.io/v1/chat/completions \ -H "Authorization: Bearer $INFERENCE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "claude-opus-4-5", "messages": [ { "role": "user", "content": "Hello, Inference!" } ], "stream": true }'Hello! I'm answering you through the Inference gateway.
Every major upstream, already wired — new models need zero code from you
Any OpenAI-compatible provider goes live as a catalog row, never a code change.
You need more than a proxy. Key governance, spend caps, automatic failover and cost observability all live in the same layer.
Fully compatible with the OpenAI Chat Completions wire format, including streaming, tool calls and vision. Your existing openai SDK, LangChain or Vercel AI SDK just points here — swap the base_url and you're done.
Stack several credentials per upstream and we rotate them by priority. On a 429, a timeout or an upstream outage the gateway retries the next healthy key inside the same request — your users never see the failure.
Scope each key to an allow-list of models, give it a monthly spend cap, tag it by environment or team, and revoke it instantly. Contractors, staging and production stay properly isolated.
Every request records token usage, cost, latency and which upstream served it. Slice by key, model or day to find the feature that is burning budget — instead of discovering one lump sum at month end.
Top up your balance through Lemon Squeezy and we deduct against real token usage in real time. No monthly fee, no per-seat pricing, no minimum spend — and credits don't expire.
No new SDK, no adapter layer, no rewriting your prompt pipeline. Request and response shapes stay exactly the same.
from openai import OpenAIclient = OpenAI( base_url="https://api.openai.com/v1", api_key=os.environ["OPENAI_API_KEY"],)# locked to a single vendor's modelsresp = client.chat.completions.create( model="gpt-5.2", messages=messages,)Locked to one vendor — every model swap is a code change
from openai import OpenAIclient = OpenAI( base_url="https://api.alphacurve.io/v1", api_key=os.environ["INFERENCE_API_KEY"],)# any model in the catalog, same callresp = client.chat.completions.create( model="claude-opus-4-5", messages=messages,)One SDK for every model, plus key governance and usage reporting
Streaming, function calling, multi-turn conversations and system prompts all keep working unchanged.
https://api.alphacurve.io/v1Every model shares one endpoint and one key. Switching is a change to the model field in your request — not a new vendor account.
| Model | Provider | Context | Input / 1M | Output / 1M |
|---|---|---|---|---|
claude-opus-4-5VisionToolsNew | Anthropic | 200K | $5.00 | $25.00 |
claude-sonnet-4-5VisionTools | Anthropic | 200K | $3.00 | $15.00 |
gpt-5.2VisionToolsNew | OpenAI | 400K | $1.25 | $10.00 |
gpt-5.2-miniToolsFast | OpenAI | 400K | $0.25 | $2.00 |
gemini-3-proVisionTools | 1M | $1.25 | $10.00 | |
gemini-3-flashVisionFast | 1M | $0.10 | $0.40 | |
deepseek-v3.2ToolsFast | DeepSeek | 128K | $0.28 | $0.42 |
grok-4.1VisionTools | xAI | 256K | $3.00 | $15.00 |
claude-opus-4-5Anthropicclaude-sonnet-4-5Anthropicgpt-5.2OpenAIgpt-5.2-miniOpenAIgemini-3-proGooglegemini-3-flashGoogledeepseek-v3.2DeepSeekgrok-4.1xAIPrices in USD, billed on actual token usage. Figures here are illustrative — the catalog API is authoritative.
See the full catalogThe dashboard aggregates token usage, cost and latency from every request in real time — sliced by key and by model, with alerts before a budget or balance runs out.
Spend this period
$48.16
+12.4%
Tokens used
34.2M
+8.1%
Requests
126,480
+3.6%
P95 latency
1.24s
-9.2%
Daily spend (USD)
Last 14 days
Cost by model
Usage by key
Top up once through Lemon Squeezy and we meter against your balance in real time. Unused credits stay yours, and you can stop any time — there is nothing to cancel.
Starter
For a proof of concept, a side project or validating the integration.
≈ 8M tokens on gemini-3-flash
Top up $10Builder
You have real users and need failover plus proper usage reporting.
≈ 17.5M tokens across a typical model mix
Top up $50Scale
Multiple environments and team keys that need spend caps and cost attribution.
≈ 71M tokens across a typical model mix
Top up $200Need a larger balance or an invoice? Talk to us
If it isn't answered here, the docs go deeper — or just email us.
Yes. We implement the OpenAI Chat Completions API, so request and response payloads are identical — including SSE streaming, function calling and multimodal input. Point base_url at https://api.alphacurve.io/v1, use an Inference key, and your SDK, prompts and post-processing stay exactly as they are.
Get started
Sign up, create your first key, and run the whole flow on trial credits before you top up. No credit card, no sales call.
Swap this one line and you're live
base_url = "https://api.alphacurve.io/v1"