Model Selection

Iris uses separate models for generation, delegated analysis, and status messages. The primary model handles commits, reviews, PRs, changelogs, release notes, and chat. Subagents use the primary model unless you configure subagent_model. The fast model writes progress messages.

Default Models and Effort

Provider	Primary model	Primary effort	Subagent effort	Status model
OpenAI	gpt-6-astra	medium	low	gpt-5.6-luna, reasoning none
Anthropic	claude-opus-5	high	low	claude-haiku-4-5-20251001, no effort parameter
Google	gemini-3.8-flash	medium	low	gemini-3.5-flash-lite

These defaults were checked against provider documentation in September 2026. Model availability still depends on your account. Existing explicit model choices stay configured until you change them. See Providers for OpenRouter and Fireworks configuration.

Configure Each Role

# Primary analysis model
git-iris config --provider openai --model gpt-6-astra

# Optional independent model for delegated analysis
git-iris config --provider openai --subagent-model gpt-5.6-terra

# Lightweight progress messages
git-iris config --provider openai --fast-model gpt-5.6-luna

The same fields work in the global provider configuration:

[providers.openai]
model = "gpt-6-astra"
subagent_model = "gpt-5.6-terra"
fast_model = "gpt-5.6-luna"

Leave subagent_model unset to use the primary model for delegated analysis. Changing fast_model only changes status generation. Use lower effort or a different worker model after comparing review findings and completion quality on representative repositories.

Upgrading Existing Configuration

Upgrades preserve explicit primary models, fast models, and provider parameters. Change saved model IDs deliberately when adopting new defaults, and review parameter overrides for compatibility with the chosen model.

Delegated analysis now uses the primary model unless subagent_model is set. Earlier versions used the fast model for workers as well as status messages. To retain a previous worker choice, set that model explicitly with git-iris config --provider PROVIDER --subagent-model MODEL. Worker cost and output quality depend on that choice; fast_model now affects status messages only.

Saved custom instructions also apply across capabilities. Scope artifact-specific rules explicitly, as described in Configuration.

Reasoning Controls

Iris chooses effort by task role. You can override provider parameters through --param:

git-iris config --provider openai --param reasoning='{"effort":"medium"}'
git-iris config --provider anthropic --param output_config='{"effort":"high"}'

Explicit parameters apply to requests using that provider configuration. Check the selected model's supported parameters before overriding defaults, especially when the status model differs from the primary model. Anthropic Haiku does not support the effort parameter.

Astra uses the Responses API for tool calling. Astra does not accept reasoning none or minimal, and does not support sampling parameters such as temperature and top_p. OpenAI migration guidance explains the request requirements.

Opus 5 uses adaptive thinking and defaults to high effort. Haiku 4.5 remains the lightweight status model. See Anthropic's model overview.

Google's Gemini 3.8 Flash supports low, medium, and high thinking. The old gemini-3-pro-preview endpoint was retired; update saved configurations that still name it. See Gemini 3.8 Flash and Google's deprecation schedule.

Context and Output Budgets

A model's context window covers input and conversation history. Its output limit is a separate constraint. A larger output budget does not make the input context window larger.

The primary Astra and Opus models have roughly one million tokens of context, as does Gemini 3.8 Flash. Haiku's context is 200K. Tool calls still use targeted file excerpts and diff summaries to keep evidence relevant. Those tools do not guarantee that every conversation fits its model's context window.

Availability and Diagnostics

Choose an explicit model ID offered by your provider. Iris reports provider errors when a selected model is unavailable; it does not silently switch the main task to a different model.

git-iris gen --debug

Debug output shows agent activity and token usage. If a request exceeds context, narrow the requested comparison or choose a model with sufficient context. Iris currently requests 16,384 output tokens for main tasks and 4,096 for subagents; token_limit is context metadata and does not change those budgets. Authentication errors require checking the active provider's key, independently of model selection.