Models
The catalog, the engine behind each id and the price per token.
Run / models
GET /v1/models needs no key and returns an OpenAI-style list. Provn adds fields for engine, context length, pricing, lane and streaming.
curl https://YOUR-PROVN-HOST/v1/models| Model | Engine | Input per 1M tokens | Output per 1M tokens | Streaming |
|---|---|---|---|---|
provn | gpt-oss-20b | $0.10 | $0.40 | Yes |
provn-nano | gpt-5.4-nano | $0.20 | $1.25 | Emulated from one upstream response |
Fields on each model
| Field | Meaning |
|---|---|
engine | The upstream model behind the id. Receipts repeat it as engine. |
context_length | The context window the gateway reports for the model. |
pricing.input_per_m_usd | USD per million prompt tokens, margin included. |
pricing.output_per_m_usd | USD per million completion tokens, margin included. |
lane | The upstream route that serves the model. |
streaming | Streaming support for the model. |
Prices
List prices include Provn's margin. The cost_micro_usd on a receipt is the amount your balance pays, and Provn adds nothing on top later. The sample receipt in these docs bills 12 prompt tokens and 48 completion tokens on provn at 21 micro-USD, or $0.000021.
The free public upstream
provn and provn-nano run on a free public upstream. Its host processes your prompts in plaintext, and Provn offers no SLA for it. GET /v1/status reports whether the gateway can reach it.
Extra models
An operator can connect an OpenAI-compatible provider by setting PROVN_UPSTREAM_BASE_URL and PROVN_UPSTREAM_API_KEY. Its models then show up in GET /v1/models on that deployment. They run on the standard privacy tier as well, so the provider sees prompt plaintext. Provn has no confidential tier.
Sandbox pricing
sandbox-js bills CPU time: $0.00005 per CPU second (50 micro-USD), with a minimum of 1 micro-USD per run.