TokenWatt API · your account

Balance and usage

Loading…

Add credit

Usage, last 30 days

Day	Requests	Input tokens	Output tokens	Cost

API key

This page is private: anyone with this link can see your balance and rotate your key. Prices: . Questions: hello@tokenwatt.io.

TokenWatt API · paused

Open models, served on GPUs we've measured.

An OpenAI-compatible API for open-weight models. Every endpoint's capacity comes from our own published benchmarks, so we admit only the traffic it can serve within its latency target and return a fast 429 instead of a slow queue. We don't log prompts or completions.

Get an API key, $1 free How this model performs

Model	Context	Input / M tok	Cached input / M tok	Output / M tok	Status
qwen/qwen3-30b-a3b-instruct-2507FP8, accuracy-checked against BF16 · tool calling, JSON output	32,768	$0.08	$0.02	$0.28	Paused

The API is paused. We ran it for a week on a measured H100 and published what we learned (capacity, cache economics, tuning). It isn't taking new sign-ups or top-ups right now. If you have a balance, your dashboard still works and unused purchased credit is refundable: write to hello@tokenwatt.io.

Cached input is 75% off, automatically. When a request repeats a prompt prefix the engine has already processed (a system prompt, tool definitions, a shared document, an agent's growing context), those input tokens cost $0.02 per million instead of $0.08. Nothing to configure: the response's usage.prompt_tokens_details.cached_tokens shows how many were cached.

Sized from measurements

We serve each model at the capacity we measured within p99 time to first token ≤ 2 s and p99 time per token ≤ 50 ms. Past that, you get an immediate 429, not a request that crawls.

Prepaid, per token

Buy credit, pay only for tokens served. Your dashboard shows balance and daily usage. When credit runs out, requests get a 402 until you top up.

No content kept

Prompts and completions are never logged or used for training. We keep only usage metadata: time, model and token counts, for billing.

Quick start

curl https://api.tokenwatt.io/v1/chat/completions \
  -H "Authorization: Bearer $TOKENWATT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen/qwen3-30b-a3b-instruct-2507",
       "messages": [{"role": "user", "content": "Hello"}]}'

Any OpenAI SDK works: set the base URL to https://api.tokenwatt.io/v1 and use your TokenWatt key. Streaming is supported and every response includes token usage.