581 models from 80 makers
One API.Every model. Best route.
One OpenAI-compatible endpoint, one key and one Toman wallet for 581 models from 80 makers. The router sends each request the best way, and the next way if one fails.
Not writing code? Try Chat
How the router chooses
router/autoAn illustration, not live data.- Route ASelected
- Route BStandby
- Route CDegraded
- 581
- models
- 80
- model makers
- 2M
- tokens, the largest context window
- 63
- models published in the last 30 days
Why teams put their traffic through us
Integrate once
An OpenAI-compatible contract that outlives the providers behind it. Point your existing SDK at us and stop maintaining five clients.
Always the best route
Every call is ranked on price, latency, region, and live endpoint health before it leaves the edge. You pick the strategy, we pick the route.
One bill, every provider
Prepaid credit or an invoiced contract. Every provider you touch lands in one usage history with per-request cost attribution.
Stays up when a provider does not
Circuit breakers pull a bad endpoint out of rotation, and safe retries move to a peer inside the same request. No 503 for your user.
What happens to a request
A request names a model, not a vendor. The router resolves that name to a ranked list of concrete endpoints, reserves credit against your wallet, and streams the first token back while the ledger settles behind it.
Read the routing docs →- 01
Authenticate and resolve
Key hash lookup, org and project scope, allowed-model check.
- 02
Filter endpoints
Drop anything failing capability, region, policy, or breaker state.
- 03
Score and rank
Balanced, cheapest, fastest, or a pinned provider order you define.
- 04
Reserve credit
Maximum-cost hold on the wallet, released and trued up on completion.
- 05
Stream, or fall back
First token wins the attempt. A pre-token failure retries the next peer.
Every model, one price list
Frontier and open-weight models, priced per million tokens.
- anthropic/claude-sonnet-5.5
- Input
- 720,776 Toman
- Output
- 3,603,880 Toman
- Context
- 1M
- Maker
- Anthropic
- openai/gpt-6.1-sol
- Input
- 720,776 Toman
- Output
- 3,603,880 Toman
- Context
- 1.1M
- Maker
- OpenAI
- google/gemini-4-argon
- Input
- 720,776 Toman
- Output
- 3,603,880 Toman
- Context
- 1M
- Maker
- deepseek/deepseek-v4.1-flash
- Input
- 20,496 Toman
- Output
- 75,152 Toman
- Context
- 1M
- Maker
- DeepSeek
- x-ai/grok-4.7
- Input
- 720,776 Toman
- Output
- 2,162,328 Toman
- Context
- 500K
- Maker
- xAI
- qwen/qwen3.8-max-0902
- Input
- 720,776 Toman
- Output
- 2,162,328 Toman
- Context
- 1M
- Maker
- Qwen
| Model | Input | Output | Context | Maker |
|---|---|---|---|---|
| anthropic/claude-sonnet-5.5 | 720,776 Toman | 3,603,880 Toman | 1M | Anthropic |
| openai/gpt-6.1-sol | 720,776 Toman | 3,603,880 Toman | 1.1M | OpenAI |
| google/gemini-4-argon | 720,776 Toman | 3,603,880 Toman | 1M | |
| deepseek/deepseek-v4.1-flash | 20,496 Toman | 75,152 Toman | 1M | DeepSeek |
| x-ai/grok-4.7 | 720,776 Toman | 2,162,328 Toman | 500K | xAI |
| qwen/qwen3.8-max-0902 | 720,776 Toman | 2,162,328 Toman | 1M | Qwen |
Free plan prices, in Toman at today's rate. Plans and prices
Newest in the catalog
The models published most recently, straight from the catalog, which follows the market several times a day.
All new models →- Thinking Machines: Inkling 2 MiniNov 21, 20261M context · 27,328 Toman/M input tokens
- RadixArk: DaisyRose 27BNov 21, 2026262K context · 54,656 Toman/M input tokens
- Perplexity: Embed V2 Context 0.6BNov 20, 202632K context · 0 Toman/M input tokens
- Perplexity: Embed V2 Context 9BNov 20, 202632K context · 0 Toman/M input tokens
- Qwen: Qwen3 TTS 1.7BNov 20, 20264K context · 0 Toman/M input tokens
Drop it into what you already wrote
Change the base URL and the key. Every official OpenAI client, LangChain, LlamaIndex, and the Vercel AI SDK work unmodified.
Use your key in Claude Code, Codex, Cursor and other tools →import os
from openai import OpenAI
client = OpenAI(
base_url="https://ai.fluxcdn.cloud/api/v1",
api_key=os.environ["FLUXCDN_API_KEY"],
)
resp = client.chat.completions.create(
model="router/auto",
messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)Enterprise
Controls your security review will actually ask about
Pin inference to a region, forbid providers that train on inputs, cap spend per project, and export an audit trail for every routing decision.
- SAML and SCIM
- Okta, Entra ID, and Google Workspace. Directory-synced roles, enforced sign-in, and automatic deprovisioning when someone leaves.
- Regional routing
- Pin a project to US-only or EU-only endpoints. Requests that cannot be served in-region fail rather than silently leaving it.
- Retention policy
- Zero-retention by default, with providers that train on inputs excluded at the org level. Enable logging per project when you need it.
- Budgets and limits
- RPM, TPM, concurrency, and daily or monthly spend ceilings per key, per project, and per organization.
- Audit logs
- Every key, member, policy, and routing-configuration change, with actor and timestamp, exportable to your SIEM.
- Invoicing
- Net-30 terms, purchase orders, committed-volume discounts, and a single monthly invoice covering every provider you route through.
Integrate once. Never migrate again.
50,000 Toman of trial credit on sign-up, 150,000 Toman more when you verify your mobile number.

