Source: Gemini models (Google AI Studio docs)
Google's Gemini API model list: Gemini 4 Carbon, Argon and Hydrogen, Gemini 3.x Flash/Flash-Lite, specialised models and deprecations.
Contents
- What: official API documentation pages (models list and the Gemini 4 Carbon model page)
- Publisher: Google
- Date: accessed 2026-11-14 (Carbon page: latest update November 2026)
- Original: Gemini models
Summary
Catalogues Gemini models by generation and status (stable, new, preview, deprecated) and links to per-model pages with token limits and capabilities.
Key claims
- Gemini 4 has three tiers: Carbon (strongest reasoning, complex SWE and multimodal knowledge work), Argon (efficient frontier reasoning, lower token spend), Hydrogen (fast, everyday coding and agents).
- Gemini 4 models take text, image, audio, video and PDF input, have a 1,048,576-token context, 65,536 default max output, and an opt-in long-decode mode up to ~1M output tokens.
- Gemini 4 Carbon supports thinking levels (low/medium/high), function calling, structured outputs, Search and Maps grounding, code execution, computer use, URL context and file search; only implicit caching.
- Gemini 3.8 Flash is labelled previous-generation; 3.5 Flash-Lite is the fastest, cheapest 3.5 model.
- Gemini 2.5 Pro and 2.5 Flash are deprecated; from 2026-12-01 requests to them are routed to 3.8 Flash and 3.5 Flash-Lite respectively.
- Specialised models include Live (voice), TTS, transcription, image (Nano Banana) and video (Omni Flash).