KYZON ROUTER

One endpoint. Every model.

A single API that routes to every major AI provider. OpenAI SDK compatible. Set team budgets. Whitelist models. Log everything. One line of code to switch.

Request Early Access Read the docs

main.py

# Switch models in one line.
# model="google/gemini-2.5-flash"
# model="openai/gpt-4.1-mini"
# model="mistral/mistral-large"

Models available

Providers supported

99.9%

Uptime SLA

<50ms

Routing overhead

WHY MULTI-MODEL

The best model depends on the task. Not the brand.

GPT-5 is not the best model for everything.

Neither is Claude or Gemini. The best model for summarisation is not the best model for code generation. Routing per task cuts cost and often improves output.

TaskTypical choiceOptimal routeCost delta

Summarisation

GPT-5 ($12/M tok)

Gemini Flash ($0.30/M)

Code generation

GPT-5 ($12/M tok)

Claude Sonnet ($3/M)

Data extraction

GPT-5 ($12/M tok)

Mistral Large ($2/M)

Customer support

GPT-5 ($12/M tok)

GPT-4.1 Mini ($0.40/M)

Complex reasoning

GPT-5 ($12/M tok)

Claude Opus ($15/M)

* Some tasks warrant the premium model. Router doesn't cut corners where quality matters. It cuts waste where quality is equivalent.

Most teams pay for the smartest model on every task.

A customer support draft doesn't need the same model that powers your core reasoning engine. One-provider routing means frontier-model prices for commodity tasks.

KYZON Router routes each request to the model that meets your quality floor at the lowest cost. You pay for capability where it matters.

HOW IT WORKS

Three lines. Not three months.

Point your SDK at Router.

Change one line in your existing OpenAI integration. Everything else stays the same.

main.py

client = OpenAI(    base_url="https://router.kyzon.com/v1"  # ← this line)

Pick the right model for the job.

Specify a model per request. Or let Router choose based on your cost and quality preferences.

routes.py

# Explicit routing — you chooseresponse = client.chat.completions.create(    model="google/gemini-2.5-flash",    messages=[{"role": "user", "content": prompt}]) # Auto routing — Router choosesresponse = client.chat.completions.create(    model="auto",  # Routes based on your policy    messages=[{"role": "user", "content": prompt}])

Set budgets. Whitelist models. Ship.

Configure team-level spend limits and model access in the dashboard. Your developers write code. Your finance team sleeps.

config.json

{  "team": "engineering",  "monthly_budget": 2000,  "models": {    "allowed": ["anthropic/*", "google/*", "openai/gpt-4.1-mini"],    "blocked": ["openai/gpt-5"]  }}

THE MATHS

Same outputs. Different invoice.

A typical SMB running 10M tokens per month across customer support, content generation, and internal tools.

BEFORE: Single provider

Single provider baseline

Customer support (4M tokens)$48.00/mo
Content generation (3M tokens)$36.00/mo
Internal tools (2M tokens)$24.00/mo
Data extraction (1M tokens)$12.00/mo

Total$120.00/mo

AFTER: KYZON Router

KYZON Router optimization

Customer support → GPT-4.1 Mini$1.60/mo
Content generation → Claude Sonnet$9.00/mo
Internal tools → Gemini Flash$0.60/mo
Data extraction → Mistral Large$2.00/mo

Subtotal$13.20/mo

Router fee (usage-based)~$1.00/mo

Total with Router$14.20/mo

MONTHLY SAVING: $105.80 (88%)

Prices based on published provider rates as of May 2026. Your actual savings depend on your usage patterns.

FEATURES

Everything you need. Nothing you don't.

OpenAI SDK compatible

Drop-in replacement. Change the base URL, keep your existing code. Works with every OpenAI SDK client library.

Intelligent routing

Set a quality floor and a cost ceiling. Router evaluates each request and routes to the model that meets both constraints.

Team budgets

Set monthly spend limits per team. When the budget runs out, requests downgrade to cheaper models or pause. No surprise invoices.

Model whitelists

Control which models each team can access. Block expensive models from development environments. Reserve frontier models for production.

Request logging

Every request logged. Model used, token count, cost, latency, team, timestamp. Exportable. Queryable via API.

Failover & retry

If a provider is down, Router fails over to an equivalent model on another provider automatically. Your application never sees the outage.

SUPPORTED MODELS

Every major provider. One endpoint.

OpenAIAnthropicGoogleMistralMetaCohereDeepSeek

120+ models across 7 providers.

Browse all models →

PRICING

Pay for what you use. Nothing else.

USAGE-BASED

Model costs pass through at provider rates. Router adds a usage fee on top. No seat fees. No minimums. No contracts.

$0–5005%

$500–2,0003%

$2,000+2%

Request Early Access

View full pricing →

Need governance, compliance, and Australian data residency? See KYZON Gateway →

ENTERPRISE

Regulated industry? Need an audit trail?

Gateway adds sovereign governance on top of Router's infrastructure. Australian data residency, PII redaction, CPS 230/234 documentation, full audit trails. Same API. Enterprise controls.

Learn about Gateway →

Start routing in 60 seconds.

Create an account, get a key, change one line. Your code stays the same. Your invoice doesn't.

Request Early Access Read the docs →