SylphxModels
27 models live · 8 providers · one Responses contract

One API for every model we sell.

Sylphx Models is a model store: pick a concrete model, send the official OpenAI Responses document, and read the answer back. Prices are published per million tokens, keys belong to your organization, and switching models never changes your client code.

Get an API keyBrowse the catalog
Base URLhttps://api.models.sylphx.ai/v1

Bearer sk-sx-… keys · revoke any key from the console

Requests · 7d
494.9K
Tokens · 7d
125.2B
Largest context
1.05M
POST /v1/responses
200 OK

Most used right now

All models and prices
{
  "id": "resp_9f2c…",
  "object": "response",
  "model": "deepseek/deepseek-v4.1-flash",
  "output_text": "Three bullets…",
  "usage": { "input_tokens": 128, "output_tokens": 96, "total_tokens": 224 }
}
DeepSeek · 1.0M context · 12.3M avg TPM487.9K requests · 7d

Models from

  • DeepSeek
  • Google
  • Meta
  • Qwen
  • Z.ai
  • xAI
  • OpenAI
  • MiniMax

Migration

Keep your OpenAI client. Point it at us.

The customer contract is the official OpenAI Responses document, so the SDK you already use keeps working. Change the base URL and the model id — nothing else moves.

  • Official Responses requests, responses, and streaming events.
  • One terminal per stream, ordered output, idempotent retries.
  • Switch models by changing one string — no per-model SDKs.
Read the quickstartOpenAPI contract
curl https://api.models.sylphx.ai/v1/responses \
  -H "Authorization: Bearer $SYLPHX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4.1-flash",
    "input": "Summarise this incident report in three bullets.",
    "stream": false
  }'

Live catalog

Prices, limits, and real traffic — in the open

Every row below is served from the same public catalog the API returns. Prices are our list prices per million tokens; usage is traffic the platform actually metered in the last seven days.

Models live278 providers
Requests · 7d494.9Kmetered on Responses
Tokens · 7d125.2Binput + output
Largest context1.05Mtokens per request

494,874 metered requests across the catalog in the last seven days.

Open the full catalog

Why Sylphx

A model shelf with the operational parts already solved

We are the seller, not a routing layer you have to reverse-engineer. What you send is what runs, what you pay is what is published, and what happened is what the response reports.

Official Responses contract

Requests and responses are the official OpenAI document. Streaming arrives as ordered events with a single terminal, so existing clients keep working when the model changes.

Request shape

Published list prices

Every model shows input, output, and cached-input prices per million tokens — the same numbers the meter bills against. No hidden tiers and no provider markup table to reverse-engineer.

See the rate sheet

Organization keys

Mint scoped sk-sx- keys for a workspace, name them per environment, watch last-used timestamps, and revoke in one click when a laptop walks away.

Open the console

Hosted tools, model-initiated

Let a model search the web, open a page, or discover a tool on its own. You pick the engine — auto, native, or a named managed engine — and constrained domains are honoured.

Tool contracts

Retries that stay honest

Send an Idempotency-Key and a retried create replays the original response instead of billing twice. Rate-limit headers and typed errors tell you exactly what to do next.

Error envelope

Data posture per model

Each row states its training posture and zero-data-retention availability, taken from the catalog, so a compliance question has a per-model answer instead of a footnote.

Check a model

Pricing

One rate sheet, our prices, no guesswork

Billing is metered per token on Responses. The numbers below are our list prices — the same values published on each model page and charged on your usage page.

Full rate sheet

Lightest per million tokens

Frontier per million tokens

BillingPer tokenmetered on every Responses call
Cache readsDiscountedrate published per model
RetriesIdempotenta replayed create is not billed twice

Console

Keys, usage, and a playground in the same place

Sign in with your Sylphx account, mint an organization key, watch what it spends, and try a model before you write code — all on the site you already trust for prices.

  • Usage you can read: requests, input and output tokens, and settled spend, broken down by model.
  • Keys with a history: creation and last-used timestamps for every sk-sx-… key, with instant revocation.
  • Playground: send a prompt to any catalog model and keep the exact request in the browser.
Create an accountOpen console
Console · Overviewworkspace

Requests

12,486

Tokens

48.2M

Spend

$356.20

Keys

●productionsk-sx-8f2a…2 minutes ago
●stagingsk-sx-1c94…yesterday

Illustrative console layout.

FAQ

Answers before you write code

Do I need a special SDK?
No. The API speaks the official OpenAI Responses contract, so the OpenAI SDKs for JavaScript and Python work after you set the base URL to https://api.models.sylphx.ai/v1 and use an sk-sx-… key. A plain HTTP client is enough too.
Which models can I call?
Whatever the public catalog lists at that moment — each row is a concrete model we sell, with its prices, context window, limits, and data posture. Request a model that is not listed and you get a typed model_not_found instead of a silent substitution.
How does billing work?
Usage is metered per token on every response and priced with the list prices published on this site. Cached input is discounted at the per-model rate shown in the catalog. Your console reports requests, tokens, and settled spend so the bill is never a surprise.
Does my traffic train someone else's model?
Each catalog row publishes its training posture and whether zero-data-retention is available, so you can choose a model that matches your policy without reading a footnote. If a model does not publish a field, we show it as unpublished rather than guessing.
Can the model search the web?
Yes, when you list the hosted tools. The model decides when a search or page open is useful, you choose the engine, and the response reports what was executed — including domain constraints you set. See hosted tools.
What happens when a request fails?
Errors come back in one typed envelope with a code, a summary, whether a retry is meaningful, and the next action. Send an Idempotency-Key and a retried create replays the original response instead of charging twice.

Your first response is one key away.

Create an account, mint an organization key, and send the official Responses document to any model in the catalog. The quickstart walks the whole path in three steps.

Get an API keyRead the quickstart

Prefer to look first? Browse the catalog — no account needed.

Sylphx Models — every model we sell, one official Responses API