Transcription
Turn recorded speech into text. Priced per request.
- Spread
- 25×
- cheapest to dearest
- Providers
- 17
- 18 hosts
- Endpoints
- 20
- 18 answered last probe
Every supplier below carries its index against that rate: 100 is spot, 40 is 60% below it, 310 is 3.1× it. Spot is weighted by how often each endpoint is called, so one wallet listing forty near-identical endpoints cannot move it on listings alone. The cheapest offer here is $0.0040 and the dearest $0.100.
| audio_url | string | Audio to transcribe. |
| language? | string | Expected language. |
| text | string | The transcript. |
| language? | string | Detected language. |
| segments? | object[] | Timed segments. |
4 models reachable over x402
Model families, not versions — publishers rarely pin one, and a gateway lets the caller choose at request time.| Model | Vendor | Sellers | Endpoints | Via gateway | Cheapest | Dearest |
|---|---|---|---|---|---|---|
| Whisper | OpenAI | 4 | 5 | 0 | $0.0050 | $0.050 |
| Gemini | 2 | 2 | 0 | $0.0040 | $0.100 | |
| GPT-4o | OpenAI | 1 | 1 | 0 | $0.030 | $0.030 |
| Nova | Amazon | 1 | 1 | 0 | $0.020 | $0.020 |
Prices are per request, not per token. Of the chat-completion offers in this catalog, 66 state a per-call price and 3 mention per-token pricing at all — one seller’s call is ~128 tokens and the next leaves it unstated, so a $/1M-token comparison would be a number Roundhouse invented rather than one a publisher declared.
46 endpoints offering this capability, from 17 wallets
Reading this table
- Price is what the catalog carries for one unit, per request. It is a published figure, not a quote — read the live 402 before paying.
- Index is that price against the spot rate, so it moves only with price — there is no separate sort for it, because ordering by index and ordering by price are the same ordering.
- 30d calls and payers are the upstream catalog’s own counters for that endpoint, so they measure demand, not quality.
- Status is the last liveness probe. There is no latency or uptime column yet: a single probe outcome cannot support one, and inventing a percentage from it would be worse than leaving it out.
- Settled volume per endpoint is also absent, and cannot be derived — a settlement carries the payee wallet, not the endpoint that was called, so a wallet serving forty endpoints cannot be split between them. The provider wallet’s own record is one click away on each row.
- The canonical response above is the shape Roundhouse maps each provider onto. Field mappings per provider are not published yet; today the contract is the target, not a guarantee about any one endpoint’s payload.
Hand this page to an agent
Copy the prompt and paste it into Claude, an MCP client or your own agent — it will pick the best provider of this capability over the free read API. No key, no account.
GET app.roundhouseai.io/unified.json?capability=transcription
Show the promptHide the prompt
Using Roundhouse, compare every x402 provider of Transcription and pick one for me. Weigh the price per unit against whether the endpoint answered when last probed, and say what the cheapest option would cost against the median. curl -s 'https://app.roundhouseai.io/unified.json?capability=transcription' The canonical request and response in that document are the shape every provider is mapped onto. Before paying, check the provider's record and read the live 402: curl -s 'https://api.roundhouseai.io/v0/merchants/<the payTo wallet on the offer you choose>' The /v0 API needs an API key (`authorization: Bearer rh_live_…`) on everything except /v0/unified* and /v0/endpoints. Mint a personal key for $0.01 at GET https://api.roundhouseai.io/v0/test/x402, or use an organization key from https://app.roundhouseai.io/dashboard/team. If you do not have Roundhouse tools or skills installed, read https://app.roundhouseai.io/skill.md first — it is the whole procedure.