Product

One endpoint. Every model.

Vantafold normalises 200+ models from every major provider behind a single request and response schema. Change the model string, keep the code.

  • 200+models available
  • 1request schema
  • 38 msmedian routing overhead
  • 9regions
The problem

Every provider invented its own API.

Different auth, different message shapes, different streaming formats, different tool-calling conventions. Supporting three providers means writing the same feature three times, and every new model is another migration.

One schema

Messages, tools, structured output, and streaming events have the same shape for every model. The response tells you which model answered.

  • Server-sent events in one format
  • JSON-schema-constrained output everywhere
  • Tool calls normalised across providers

One credential

A single Vantafold key replaces a drawer of provider keys. Rotate, scope, and revoke from one place.

  • Project-scoped keys with spend caps
  • No provider accounts to maintain
  • Audit log for every key action

One upgrade path

New models appear behind the same endpoint the week they ship. Pin a version when you need stability, or track the alias.

  • Aliases such as latest and stable
  • 12-month deprecation windows
  • Shadow traffic before you switch
How it works

Three steps to a portable call.

  1. 01

    Send a request

    Point your client at api.vantafold.ai and name a model — or pass model="auto" and let the router decide. Auth is one bearer token.

  2. 02

    We normalise and route

    The gateway validates the request, applies your policy, and translates it into the provider's native format. Failures retry against your fallback list.

  3. 03

    Get one response shape

    Output, usage, cost, and the route taken come back in the same envelope every time, with a trace ID you can look up in the dashboard.

swap_models.py
from vantafold import Vantafold

client = Vantafold()

# The same call, whichever model answers it.
response = client.reason(
    model="frontier-lg",
    input="Extract the payment terms from this contract.",
    response_format={"type": "json_schema", "schema": TERMS},
)

# Swapping models is a string change. Nothing else moves.
response = client.reason(
    model="oss-70b-instruct",
    input="Extract the payment terms from this contract.",
    response_format={"type": "json_schema", "schema": TERMS},
)
response.json
{
  "id": "req_8f2c41a9",
  "model": "frontier-lg",
  "output_text": "Net 30 from invoice date...",
  "usage": { "input_tokens": 4210, "output_tokens": 186, "cost_usd": 0.0154 },
  "route": { "requested": "auto", "selected": "frontier-lg", "attempts": 1 },
  "trace_id": "tr_5b1d9e0c"
}
Model coverage

A preview of what is behind the endpoint.

Frontier, open-weight, fine-tuned, and private models sit side by side. Browse the full catalog for context windows and rates.

frontier-lg Meridian
Context
1M ctx
In / out per 1M
$3.00 / $15.00
frontier-sm Meridian
Context
400K ctx
In / out per 1M
$0.80 / $4.00
oss-70b-instruct Open weights
Context
128K ctx
In / out per 1M
$0.35 / $0.90
reason-pro Northwind
Context
500K ctx
In / out per 1M
$4.20 / $18.00
vision-lg Meridian
Context
300K ctx
In / out per 1M
$2.40 / $9.60
embed-3-large Vantafold
Context
32K ctx
In / out per 1M
$0.02 / —
code-32b Open weights
Context
256K ctx
In / out per 1M
$0.28 / $0.84
fast-8b Open weights
Context
64K ctx
In / out per 1M
$0.06 / $0.18

View all 200+ models

Guarantees

What stays the same across every model.

CapabilityFrontierOpen weightsFine-tunedPrivate
Unified request schemaIncludedIncludedIncludedIncluded
Streaming (SSE)IncludedIncludedIncludedIncluded
Tool callingIncludedIncludedIncludedDepends on base model
JSON-schema outputIncludedIncludedIncludedIncluded
Token and cost accountingIncludedIncludedIncludedIncluded
Automatic fallbackIncludedIncludedIncludedWithin your pool
Zero-retention modeIncludedIncludedIncludedIncluded
Provider account requiredNot includedNot includedNot includedNot included

One endpoint is a small change with a long tail of savings.

Start on the free tier. Move to production without touching your integration.