today
The token mill is running
Cheap Qwen3 has arrived at the market. Requests are leaving the station on schedule.
Fast inference.
Small-town service.
Open-weight models, an OpenAI-compatible API, and cost-aware routing across the best providers. Pay per call with x402, or keep a credit balance.
curl https://api.inference.town/v1/chat/completions \
-H "Authorization: Bearer $INFERENCE_TOWN_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3-235b-a22b-2507",
"stream": true,
"messages": [{"role":"user","content":"Hello!"}]
}'That is the whole migration from the OpenAI SDK: new base URL, new key.
The token mill is running
Today's market
| Model | Input | Output | $/1M |
|---|---|---|---|
| DeepSeek V3.2 | $0.26 | $0.38 | Sale |
| Qwen3 235B A22B | $0.09 | $0.55 | Sale |
| GPT-OSS 120B | $0.03 | $0.17 | Cheap |
Town news
The machinery behind the square
The town is not decoration. Every building maps to something real in the request path.
Router station
Cost-aware routing
Each request scores providers on price, latency, reliability and cache affinity, then keeps fallbacks ready. If a provider stalls before the first token, the train is rerouted.
Payment booth
Agent-native payments
x402 lets a software agent pay for a top-up without a human checkout. Funds become credit; every call is metered locally and settled in batches, never per request.
GPU forge
Measured, not promised
We record token counts, cost, TTFT and throughput for every call. Margin floors stop the router from selling below cost, and the town only shows aggregate, anonymized activity.