Every id, on its own plate

The models this gateway serves right now: the engine behind each id, the context window, the price you pay per million tokens and the privacy tier it really runs under. The list comes from the running gateway configuration.

provn

engine gpt-oss-20b

flagship

The stable default id. The engine behind it is named on every receipt.

Output price: $0.40 / 1M
Context window: 131,072 tokens
Input: $0.10 / 1M
Streaming: native
Tier: standard

provn-nano

engine gpt-5.4-nano

fast

Short answers and routing decisions. Streams are replayed from one upstream response.

Output price: $1.25 / 1M
Context window: 400,000 tokens
Input: $0.20 / 1M
Streaming: emulated
Tier: standard

sandbox-js

engine quickjs-wasm

JavaScript in a fresh QuickJS WebAssembly VM per run. 32 MB memory, up to 5 s of CPU, no network, no filesystem. Billed per CPU millisecond.

$0.00005 per CPU second

Sandbox docs

Standard tier only

The upstream host reads prompts in plaintext. Provn stores none of it, but it runs no confidential tier and does not pretend to.

The id is an indirection

provn is a stable name. The engine behind it can change, and every receipt names the engine that served the call.

No scores, no ranks

Bars on these plates show price and context length. They rate nothing, because Provn has no evidence for a quality ranking.