AsyncRouter · Coming soon

One endpoint. Every model, every provider.

A routing service for language models, not something you install. Your application calls one endpoint with one key, and the router serves the model you asked for from a healthy provider. GPT, Claude, Grok, Gemini and the best open models, with automatic failover and the cost of every call in one place.

Optimise for
HostLatencyThroughput$/1MUptime
openairouted420ms48 t/s$50.0099.90%
azure-openai460ms44 t/s$50.0099.95%
aws-bedrock510ms40 t/s$52.0099.80%
Request
curl https://api.asyncrouter.co.za/v1/chat/completions \
 -H "Authorization: Bearer $ASYNC_API_KEY" \
 -d '{"model":"gpt-6-astra","route":"price"}'

Illustrative figures. Providers are the routing targets.

The routing surface, mocked locally. Host names and figures are illustrative placeholders, but the behaviour they demonstrate is the real idea: ranking, degraded pass-over, failover.

Providers of every model

  • OpenAI
  • AWS
  • Anthropic
  • Microsoft Azure
  • xAI
  • Meta
  • Google Cloud
  • Mistral AI
  • DeepSeek
  • Qwen
  • Nvidia
  • Hugging Face
  • Perplexity
  • Alibaba Cloud
  • Moonshot AI
  • Databricks
  • Replicate

Provider marks belong to their respective owners

Model lineup

Top models from every lab, served from day one.

One key, one endpoint, every family below routed across healthy hosts. Pick a card to tell us which ones you would actually call.

GPT-6 Astra

Frontier

OpenAI’s newest flagship. The strongest reasoning you can route, priced like it.

Hardest tasksAgentsCode

from $50.00/1M · 420ms · 1M

Claude Opus 5

Frontier

Anthropic’s top model. Superb taste in code, writing, and long complex work.

CodeWritingComplex work

from $25.00/1M · 380ms · 1M

Grok 4.6

Frontier

xAI’s flagship for long-running agents, with real-time instincts.

AgentsResearchValue

from $6.00/1M · 290ms · 500K

Gemini 3 Pro

Frontier

Google’s best. Huge context and native multimodality for documents and media.

Long contextMultimodalDocs

from $12.00/1M · 250ms · 1M

GPT-5.6 Luna

Efficient

The lean OpenAI tier. Near-flagship quality for everyday volume.

High volumeEverydayCost

from $6.00/1M · 150ms · 400K

Llama 3.3 70B

70B

The dependable open-weight workhorse — strong instruction-following for everyday workloads.

General chatSupportDrafting

from $0.42/1M · 180ms · 128K

Mistral Small 3

24B

Small, fast, and cheap — the default for high-volume, latency-sensitive calls.

High volumeLow latencyCost

from $0.14/1M · 120ms · 32K

Qwen2.5 72B

72B

A heavyweight for multilingual and technical work, with deep context to match.

MultilingualTechnicalLong context

from $0.45/1M · 260ms · 128K

DeepSeek V3

671B MoE

Frontier-grade reasoning and code at open weights — the value pick for hard problems.

ReasoningCodeValue

from $0.35/1M · 320ms · 64K

Qwen3 32B

32B

A mid-size all-rounder with a very long memory — chat, tools, and multilingual work.

ChatTool useMultilingual

from $0.18/1M · 175ms · 256K

Gemma 3 27B

27B

Top-tier quality that fits on a single GPU — fast, cheap, and multimodal.

SpeedVisionSingle-GPU

from $0.12/1M · 140ms · 128K

Figures are illustrative placeholders. Providers are the routing targets.

How it works

Three steps between your app and every model.

01

Point your app at one endpoint

One URL, one key, the API shape your SDKs already speak. No per-provider wiring, no keys scattered through env.

02

Say what to optimise for

Price, latency, or availability, set per call. Change your mind any time; the route decision is made fresh on every request.

03

The router serves and fails over

It picks a healthy host for the model you asked for. If that host degrades mid-stream, the call moves, and you see what happened and what it cost.

  • One endpoint across model families and providers, speaking the API your SDKs already use
  • Frontier and open models side by side, so you can mix quality and cost per call
  • Route by price, latency, or availability, with failover when a provider degrades
  • Per-call cost and latency visible across every model you use

Serve models through one endpoint?

Tell us which models you use and what hurts most: price, latency, or reliability. We will come back when there is an endpoint to call.