NNyquest docs

Picking a Model

Nyquest provides a live catalog of models across multiple providers. Compare their capabilities, input/output prices, and availability for your task.

Where the model picker lives

The model picker is in the top bar, showing the currently active model name. Click it to open. You can:

  • Type to search by name or provider
  • See per-token cost (input + output, per million tokens)
  • See speed indicators
  • See which providers you have a BYOK key for (those have a green badge)

Your selection applies to the current conversation. New chats start back on Auto — the router — so a one-off experiment never silently becomes your default. Prefer a specific model? Pick it at the start of each chat.

Choosing for your task

Start with Auto or use the live catalog to compare currently available models. Check tool support for Agent Mode, image-input support for vision, and the published context window for long documents. Availability, context sizes, and provider prices can change.

Cost in practice

Check input and output rates separately in the picker or pricing catalog. A tier name is not a fixed price. Cost depends on context sent, generated tokens, the model, and eligible compression credits. Agent Mode can make several calls; Studio media is billed at provider cost. Review Usage for settled charges.

BYOK vs platform-hosted

BYOK (your key)Platform-hosted (our keys)
CostYour provider charges you directly; Nyquest is freeWallet settlement at catalog rates, with eligible compression credits
SetupAdd and activate the key in Settings → My ProvidersNothing — just pick a model
Best forHeavy users, those with provider creditsTrying things, light usage

You can mix: BYOK for OpenAI, platform-hosted for Anthropic, etc. The picker shows which is active per-model.

How to compare models for your use case

The honest answer: just try them. Send the same prompt to 3 different models and compare. The picker remembers each model's last conversation, so flipping back and forth is fast.

If you want a more rigorous comparison: open Agent Mode, give it a goal, then re-run with a different model. Agent runs are reproducible enough for casual benchmarking.

Where to next