- Context
- 131k
- Input /M
- $0.0735/M
- Output /M
- $0.294/M
- TTFT
- N/A
- Throughput
- N/A
TokenWatt
TokenWatt publishes, and confirmed in writing, that prompts and completions are processed in memory only, never stored or logged, and never used for training. Request metadata (IDs, timestamps, model, status, token counts, latency) is kept for 30 days. Inference runs in US data centers.
Privacy can differ by model. See each model’s label.
Models served by TokenWatt
Prices are per million tokens. TTFT is the observed time to first token. Selectable routes include the applicable provider-selection markup; fixed routes show the standard model price. Availability and pricing refresh continuously; each model page shows the live provider comparison.
Using TokenWatt via the API
Call a model served by TokenWatt using its displayed model ID, including the provider suffix when selection is supported.
cURL
curl https://api.nano-gpt.com/api/v1/chat/completions \
-H "Authorization: Bearer $NANOGPT_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3-coder-30b-a3b-instruct:tokenwatt",
"messages": [{"role": "user", "content": "Hello"}]
}'For models with provider selection, append :tokenwatt to the model ID, set "provider": "tokenwatt" in the request body, or send an X-Provider: tokenwatt header. Explicit provider selection adds a route-specific markup over that provider's base price; the prices in the table above already include it. Models without provider selection use the displayed model ID without a provider suffix or selection surcharge.
See the API documentation for provider preferences, price-aware routing, and error behavior.