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.
| Host | Latency | Throughput | $/1M | Uptime |
|---|---|---|---|---|
| openairouted | 420ms | 48 t/s | $50.00 | 99.90% |
| azure-openai | 460ms | 44 t/s | $50.00 | 99.95% |
| aws-bedrock | 510ms | 40 t/s | $52.00 | 99.80% |
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
Provider marks belong to their respective owners
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
FrontierOpenAI’s newest flagship. The strongest reasoning you can route, priced like it.
from $50.00/1M · 420ms · 1M
Claude Opus 5
FrontierAnthropic’s top model. Superb taste in code, writing, and long complex work.
from $25.00/1M · 380ms · 1M
Grok 4.6
FrontierxAI’s flagship for long-running agents, with real-time instincts.
from $6.00/1M · 290ms · 500K
Gemini 3 Pro
FrontierGoogle’s best. Huge context and native multimodality for documents and media.
from $12.00/1M · 250ms · 1M
GPT-5.6 Luna
EfficientThe lean OpenAI tier. Near-flagship quality for everyday volume.
from $6.00/1M · 150ms · 400K
Llama 3.3 70B
70BThe dependable open-weight workhorse — strong instruction-following for everyday workloads.
from $0.42/1M · 180ms · 128K
Mistral Small 3
24BSmall, fast, and cheap — the default for high-volume, latency-sensitive calls.
from $0.14/1M · 120ms · 32K
Qwen2.5 72B
72BA heavyweight for multilingual and technical work, with deep context to match.
from $0.45/1M · 260ms · 128K
DeepSeek V3
671B MoEFrontier-grade reasoning and code at open weights — the value pick for hard problems.
from $0.35/1M · 320ms · 64K
Qwen3 32B
32BA mid-size all-rounder with a very long memory — chat, tools, and multilingual work.
from $0.18/1M · 175ms · 256K
Gemma 3 27B
27BTop-tier quality that fits on a single GPU — fast, cheap, and multimodal.
from $0.12/1M · 140ms · 128K
Figures are illustrative placeholders. Providers are the routing targets.
Three steps between your app and every model.
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.
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.
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.