Skip to main content
Gloo AI gives you leading models from multiple providers. The Model Explorer in Gloo Studio compares them side by side, including reasoning capabilities, modalities, and speed ratings. Not sure which to pick? Cost and Speed in the AI Learning Center walks through choosing one. To match a model to your endpoint, see Choosing a model. The catalog below is fetched live in your browser from the public, unauthenticated GET /platform/v2/models endpoint, so it always reflects the current catalog. Use the Model ID as the model parameter on the endpoint for that model type. Generation models work with Responses and Completions; models marked Embedding work with the Embeddings API.
The rates shown are list prices. Direct endpoints, including Embeddings, bill at the published rate; guarded generation endpoints are priced at that rate plus 6.5%, as described in Studio Billing. Track real spend in the Gloo Studio billing dashboard.

Model catalog

Generation model IDs work across the direct Direct Responses and Direct Completions endpoints, and the guarded Responses API and Completions V2. Claude model IDs, except Claude Fable models, also work with Anthropic Messages. Embedding model IDs are for the Embeddings API. Capability flags (supports_tools, supports_streaming, supports_reasoning, supports_vision) apply to generation; the model catalog returns them for all entries. Image input (vision). A model with supports_vision: true reads images you send. Every Claude, Gemini and Grok chat model does, as does every OpenAI GPT model except the open-weight GPT OSS models, and many open-source models. See Which Models Read Images, the Understand Images with Claude tutorial and Images in Conversations. Gloo AI can route generation requests to providers that serve the model. Embedding requests use the embedding model you name and are not failed over to another provider. See Provider Failover.

Reasoning

Each model in GET /platform/v2/models has a reasoning object, or null when the model doesn’t reason. These fields describe the reasoning settings on the Responses and Chat Completions endpoints; see Reasoning for how to pick an effort level, and Models That Think Before They Answer for what reasoning is. On Anthropic Messages, Claude models use Anthropic’s own thinking parameter instead, with their own defaults. A model’s max_output_tokens is the most it can produce in one response, reasoning and answer together. To size your own limit for a reasoning model, see Set your output limit for reasoning models. In the catalog, supports_reasoning means the model reasons, whether or not Gloo lets you control how much. The catalog’s Effort levels column shows the effort_levels field described above. reasoning_level is an overall quality rating from 0 to 5, not a measure of reasoning. See GET /platform/v2/models for every field, and Reasoning for how to pick an effort.

Deprecated Models

In the catalog table above, a deprecated model shows a Deprecated badge next to its ID, with its deprecation note inline. These come from the is_deprecated and deprecation_note fields on GET /platform/v2/models, alongside replacement_model, the ID that requests are routed to. For generation models, where a replacement has been chosen, requests using the deprecated ID may continue to be routed to that replacement while you migrate. Do not assume replacement routing applies to embeddings: changing embedding models requires re-embedding stored content. Read deprecation_note for the specifics, including any price change or missing replacement. When nothing replaces a model, requests are refused with 410 and model_retired from its retirement date; see When nothing replaces a model. See Model Lifecycle for how a replacement is chosen, what the price can do, how much notice you get, and how to detect all of it from the API. Model Updates lists every model added, retired or redirected, including retired models no longer in this table.

Prompt Caching

For generation models, the Caching column shows which models support prompt caching, and which type — Implicit (automatic, e.g. OpenAI, DeepSeek, Gemini, Qwen) or Explicit (opt-in per request, e.g. Anthropic). Caching does not apply to embedding models. For each provider’s caching mechanism, billing rates, and best practices, see Prompt Caching.

Next steps

Model Lifecycle

How models are introduced and retired.

Model Updates

Every model added, retired or redirected, with dates.

Reasoning

Reasoning effort and summaries by model family.

Responses

Use a model with the Responses API.

Completions V2

Pin a model or let Gloo route.