Models and model selection

Open the model picker#

The Chat composer model control is the source of truth for account and route availability. Selecting the model name beside the composer controls opens the picker. The picker groups rows by model family, supports provider filters and search, and opens an information view with the provider, release label, context and output fields, published prices when present, and capability indicators. A capability describes the selected catalog route; it does not guarantee identical behavior on every provider route.

Capabilities and access costs#

Each table lists one model per row, grouped alphabetically by provider and model name. The icons describe the published catalog capabilities. ZDR means zero data retention. Privacy policies can differ between routes to the same model; an asterisk identifies policies available on only some routes.

Access cost shows the published upstream API rates used to access these models, in US dollars. Token rates are per one million tokens; audio duration and character rates show their own units. These are model access rates, separate from AI Integrator subscription prices. Tiered context pricing, routing, caching, and additional tool charges can change the cost of a request. Open More rates for published cache and tier details.

Rates and capabilities come from the AI Gateway model catalog, refreshed hourly. A missing rate is marked Not published. Default identifies the application's configured default for that model type; account settings can override it.

Language models (LLM)#

Language models power Chat conversations, including the Claude, GPT, Gemini, and other families listed below. The model picker determines availability for the account and its privacy settings. Installed Code runtimes have their own model selections.

Speech to text (STT)#

Transcription models turn recorded speech into text. Choose a speech-to-text model in Settings → Speech & Voice. The Live audio icon identifies a streaming audio route in the catalog; each application's recording controls determine how it is used.

Text to speech (TTS)#

Speech models generate audio for read aloud. Choose the model and an available voice in Settings → Speech & Voice. Character-based rates apply to the input text.

Embedding models#

Embedding models support retrieval across memory, conversation history, and indexed documents. Their rates apply to the text being embedded. Changing the embedding model can require existing content to be indexed again; see Memory.

Reranking models#

Reranking models score retrieved results for relevance. These catalog entries are separate from the Chat model picker; inclusion in the catalog does not mean every retrieval uses a reranker.

Reasoning and privacy locks#

The reasoning slider folds down to None for models without reasoning support. Privacy filters can require Zero Data Retention or no training; rows without a compatible route are locked. Premium rows can require Pro or Max. Fast is a separate route when the catalog publishes one and can have different metering. See Usage, Privacy, Voice, and Settings.