Models & picking one
A single key gives you access to 52 models. To switch models, just change the model field — no reconnecting needed.
Which model to use
| Task | Model | Why |
|---|---|---|
| Coding and agents | gpt-5.6-sol | strong reasoning and reliable tool calling |
| Cheap and high-volume | deepseek-v4-flash | low consumption coefficient |
| Long context, PDFs, images | gemini-3.7-flash | multimodal input, huge context window |
| Writing and analysis | claude-sonnet-4-6 | quality prose, long documents |
| Fast chat replies | gpt-5.6-luna | low latency at a low price |
| Speech recognition | gpt-4o-transcribe | audio to text |
A good strategy is to keep two models on hand: a strong one for complex steps and a cheap one for routine tasks. Switching is a one-line change.
Popular models
| Model | Context | Multiplier |
|---|---|---|
gpt-6.1-sol | — | ×3 |
gpt-6-astra | 1M | ×7.5 |
gpt-6-sol | 1M | ×2 |
gpt-6-luna | 1M | ×0.25 |
gpt-5.6-sol | 1M | ×3 |
gpt-5.6-terra | 1M | ×1.5 |
gpt-5.6-luna | 1M | ×0.22 |
gpt-5.5 | 400K | ×3 |
gpt-5.4 | 1M | ×1.5 |
gpt-5.4-mini | 400K | ×0.8 |
claude-fable-5-1 | 200K | ×8 |
claude-fable-5 | 200K | ×8 |
The full catalog with multipliers, context windows, and supported input types is on the models page.
Listing models via request
You can fetch current model names straight from the API — handy for select boxes in your UI.
curl https://api.relaymodels.com/v1/models -H "Authorization: Bearer $RELAY_API_KEY"Fallback model
If uptime matters to you, add a simple safety net: retry the request on another model when one fails.
const chain = ["gpt-5.6-sol", "claude-sonnet-4-6", "gemini-3.7-flash"];
for (const model of chain) {
try {
return await client.chat.completions.create({ model, messages });
} catch (e) {
// try the next model
}
}