Models

Fresh

Antigravity provides access to multiple reasoning models. Model selection is sticky per conversation turn -- the model you select remains active until you change it.

Primary Reasoning Models

These models are selectable by the user for agent conversations:

Model	Tier
Gemini 3.1 Pro (high)	High-capability reasoning
Gemini 3.1 Pro (low)	Faster, lower-cost reasoning
Gemini 3 Flash	Speed-optimized
Claude Sonnet 4.6 (thinking)	Anthropic reasoning model with extended thinking
Claude Opus 4.6 (thinking)	Anthropic's most capable model with extended thinking
GPT-OSS-120b	Open-source 120B parameter model

Supplementary Models

These models are used internally by Antigravity for specific tasks and are not user-selectable:

Model	Purpose
Nano Banana Pro 2	Image generation
Gemini 2.5 Pro UI Checkpoint	Browser subagent operations
Gemini 2.5 Flash	Checkpointing and summarization
Gemini 2.5 Flash Lite	Codebase search indexing

TIP

Choose your model based on the task complexity. Use high-tier models for architecture decisions and complex refactors. Use Flash or low-tier models for simple edits and quick tasks.