Model selection
Environment variable | CLI equivalent | Description |
|---|---|---|
JUNIE_MODEL | --model | LLM model to use. If not specified, Junie uses the |
JUNIE_LLM_PROVIDER | --provider | BYOK provider to use. Forces Junie to use a specific BYOK provider. Possible values: |
JUNIE_EFFORT | --effort | Reasoning effort level for models that support it. Possible values: |
To see all available models, run junie --help or use /model in interactive mode. To change the effort level during a session, use /model or the dedicated /effort command.
LLM providers
Junie supports four types of LLM providers:
Junie — models accessed through a JetBrains AI subscription, by means of a Junie login or a JetBrains AI API key. No additional configuration is required.
BYOK (Bring Your Own Key) — models accessed using your own API key from a third-party provider such as OpenAI, Anthropic, Google, xAI, or OpenRouter. See BYOK for setup instructions.
Custom — models defined in a custom profile JSON file. See Custom LLM models for details.
Proxy — models accessed through a custom proxy endpoint configured in
config.json.
When a model is available through multiple providers, Junie prefers the Junie provider by default. Use --provider to override this behavior.
Provider auto-detection
When you specify --model without --provider, Junie resolves the provider automatically:
If the Junie provider is available (you are logged in with a JetBrains account or API key), Junie uses it by default.
If the Junie provider is not available (no login or API key), Junie looks through the connected BYOK providers and selects the first one that offers the requested model.
For example, running junie --model grok --grok-api-key <key> without a JetBrains login will automatically route to the xAI BYOK provider because it is the only available provider with the Grok model.
How Junie uses models internally
The model you select (or the Default) is your primary model — Junie uses it for the main reasoning and code generation work.
However, Junie also uses a separate model for certain internal tasks such as:
Summarizing context
Classifying and routing tasks
Extracting information for memory
Filtering capabilities
Other helper tasks
These internal tasks don't need the full power of the primary model, so Junie automatically picks an alternative from the same provider. Note that these models are not always smaller — for example, routing may use GPT-4.1 by default. If your primary model is from Anthropic, the helper model will typically be Claude Haiku; if it's from Google, it will be Gemini Flash.
This applies even when using BYOK. If you bring your own API key, Junie will still use a separate model from the same provider for these internal tasks. This means you may see API calls to a model you didn't explicitly select — that's expected and helps keep costs down and responses fast. For more details on configuring custom LLM providers, see Custom LLM Models.
Junie CLI also supports model aliases. Each alias maps to a specific model version set by the JetBrains team based on our own internal evaluations and is not necessarily the latest available version. The model behind each alias may change as new models are released.
Model ID | Provider | Current model |
|---|---|---|
| Anthropic | |
| Anthropic | |
| OpenAI | |
| OpenAI | |
| ||
| ||
| xAI |