The Image input column shows native image support. A model marked — can still work with images if you set a fallback vision model — CTOR describes each image as text so the model can use it.
Anthropic
Anthropic supports two authentication modes: an API key (pay-per-token billing) or Login with Claude OAuth (credit-based usage against a Claude Pro or Max subscription). Configure either in Settings > Providers > Anthropic. See settings for details.
OpenAI
gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna support speed mode — a Normal/Fast toggle appears in the input bar when one of these models is selected.
OpenAI Codex
This is the OpenAI Codex API as a model provider for the built-in agent. To run the actual Codex CLI as your chat’s agent instead, see CLI Agents.
gpt-5.3-codex-spark requires a ChatGPT Pro plan. It appears in the model selector only when the signed-in OpenAI Codex account is on the Pro plan.
gpt-5.6-sol is unavailable on the ChatGPT Free plan. It appears in the model selector only when the signed-in OpenAI Codex account is on a Go, Plus, Pro, Business, or Enterprise plan.
gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, and gpt-5.4 support speed mode — a Normal/Fast toggle appears in the input bar when one of these models is selected.
xAI
composer-2.5-fast requires a SuperGrok subscription — sign in with Login SuperGrok in settings. It does not work with a plain xAI API key.
On a fresh install with no saved model choice, both Google providers start on
gemini-3.7-flash.
Google Agent Platform (Vertex AI)
Google rebranded Vertex AI to Agent Platform; CTOR shows both names so the provider is findable by either label. Gemini models go through the Generative AI API on Vertex; Claude models use Anthropic’s Vertex partner endpoint. Authentication uses Application Default Credentials (ADC), a service account key file, or an API key (Gemini only — Claude partner models reject API keys). See settings for setup details.
GitHub Copilot
GitHub Copilot uses OAuth for authentication. Configure it in Settings > Providers > GitHub Copilot.
OpenAI Codex uses OAuth for authentication (Login with OpenAI). Configure it in Settings > Providers > OpenAI Codex.
OpenCode Zen
OpenCode Go
OpenAI Compatible endpoints
Add any number of OpenAI-compatible API endpoints, each with its own name, base URL, API key, and model list. Manage endpoints in Settings > OpenAI Compatible Endpoints. Each endpoint is configured with:
Models from configured endpoints appear in the model selector grouped under the endpoint name. This allows you to use local models (Ollama, LM Studio) or any third-party provider that implements the OpenAI API format.
Per-model options
Each model you add to an endpoint carries its own options:
The reasoning options unlock in sequence: set the endpoint API Mode to Responses to enable Reasoning, then enable Reasoning to unlock XHigh and Max.
Thinking levels
Models with reasoning support can use different thinking levels to control the depth of analysis:
Extra High and Max are only available on models marked accordingly in the extended thinking column above. A few models — claude-sonnet-4.6, claude-opus-4.6, glm-5.3, glm-5.2, deepseek-v4-pro, and deepseek-v4-flash — expose Max without Extra High, jumping straight from High to Max.
Ultra
Ultra is an experimental tier that builds on Max. It sends the same reasoning budget as Max, and adds multi-agent orchestration: the agent proactively fans work out to parallel sub-agents and merges their reports back into the conversation. Expect the highest token usage of any tier. Turn it on under Settings → System → Experimental features → Ultra Thinking Level. While the setting is off, Ultra does not appear in the picker, and a session that previously used it runs as plain Max with no orchestration. Ultra is chosen per session rather than globally — it is not offered as a default thinking level, and scheduled runs fall back to Max so nothing fans out unattended. Only these models offer Ultra:
See Sub-agents for what the agent does with the tier.
Speed mode
When using a supported OpenAI or OpenAI Codex model, a speed mode selector (gauge icon) appears to the right of the thinking level selector in the input bar.
Fast maps to the OpenAI
service_tier: 'priority' parameter. The setting is persisted per session.
The selector is only visible when a model that supports speed mode is active. Switching to a model without speed mode support hides the control; any previously set “Fast” preference is not applied.
Environment variable fallback
If no API key is configured in the app settings, the agent checks these environment variables:ANTHROPIC_API_KEY— used for Anthropic modelsOPENAI_API_KEY— used for OpenAI models