Skip to content

Gemini models

Google's Gemini models in the Gemini API: Gemini 4 Carbon, Argon and Hydrogen, the 3.x Flash models, and shutdown dates for 2.5.

  • API only
  • Models 11
  • Context up to 1M
  • Updated Nov 2026
Model docs Compare
On this page

At a glance

Type
API-only models
Developer
Google DeepMind (United States)
First release
Dec 2023
Models tracked
11 (4 current)
Longest context
1M tokens
Abilities
Tool callingReasoning modeImage inputAudio input
Access
Proprietary. Available through the Gemini API (Google AI Studio) and Vertex AI. For downloadable Google models, see Gemma.
Website
deepmind.google

Newest model

Gemini 4 Hydrogen

Released Nov 2026 · 1M context

Next retirement

Gemini 2.5 Pro

1 Dec 2026 · in 17 days

All Gemini models

ModelReleasedAPI IDContextAbilitiesStatus
Gemini 4 Hydrogen Everyday coding and agents; up to 1M output tokensNov 2026 gemini-4-hydrogen1MToolsReasoningVisionAudio Current
Gemini 4 Carbon Strongest reasoning: software engineering, research, securityNov 2026 gemini-4-carbon1MToolsReasoningVisionAudio Current
Gemini 4 Argon Efficient frontier reasoning, faster and cheaper than CarbonOct 2026 gemini-4-argon1MToolsReasoningVisionAudio Current
Gemini 3.5 Flash-Lite Fastest, cheapest model for high throughputJul 2026 gemini-3.5-flash-lite1MToolsReasoningVisionAudio Current
Gemini 3.1 Pro (preview)Feb 2026 gemini-3.1-pro-preview1MToolsReasoningVisionAudio Preview
Gemini 3.8 Flash Previous-generation Flash for agents and enterprise workflowsSep 2026 gemini-3.8-flash1MToolsReasoningVisionAudio Previous generation
Gemini 3.6 FlashJul 2026 gemini-3.6-flash1MToolsReasoningVisionAudio Previous generation
Gemini 3.1 Flash-LiteMay 2026 gemini-3.1-flash-lite1MToolsReasoningVisionAudio Previous generation Retires 7 May 2027
Gemini 2.5 ProJun 2025 gemini-2.5-pro1MToolsReasoningVisionAudio Deprecated Retires 1 Dec 2026
Gemini 2.5 FlashJun 2025 gemini-2.5-flash1MToolsReasoningVisionAudio Deprecated Retires 1 Dec 2026
Gemini 2.5 Flash-LiteJul 2025 gemini-2.5-flash-lite1MToolsReasoningVisionAudio Legacy

Sizes, context windows and abilities come from the vendor's model documentation, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

Gemini lifecycle and retirement dates

Gemini 2.5 ProDeprecated

Retires 1 Dec 2026 · in 17 days

Move to Gemini 3.8 Flash (automatic reroute)

Gemini 2.5 FlashDeprecated

Retires 1 Dec 2026 · in 17 days

Move to Gemini 3.5 Flash-Lite (automatic reroute)

Gemini 3.1 Flash-LitePrevious generation

Retires 7 May 2027 · in 5 months

Move to Gemini 3.5 Flash-Lite

ModelReleasedStatusRetirementReplacement
Gemini 4 Hydrogen12 Nov 2026CurrentNot announced—
Gemini 4 Carbon10 Nov 2026CurrentNot announced—
Gemini 4 Argon13 Oct 2026CurrentNot announced—
Gemini 3.5 Flash-Lite21 Jul 2026CurrentNot announced—
Gemini 3.1 Pro (preview)19 Feb 2026PreviewNot announced—
Gemini 3.8 Flash2 Sep 2026Previous generationNot announced—
Gemini 3.6 Flash21 Jul 2026Previous generationNot announced—
Gemini 3.1 Flash-Lite7 May 2026Previous generation7 May 2027Gemini 3.5 Flash-Lite
Gemini 2.5 Pro17 Jun 2025Deprecated1 Dec 2026Gemini 3.8 Flash (automatic reroute)
Gemini 2.5 Flash17 Jun 2025Deprecated1 Dec 2026Gemini 3.5 Flash-Lite (automatic reroute)
Gemini 2.5 Flash-Lite22 Jul 2025LegacyNot announced—

Source: official deprecation page, checked 14 Nov 2026. Dates apply to the vendor's own API; cloud partners (AWS Bedrock, Google Vertex AI, Azure) can set their own schedules.

What Gemini is good for

Video and audio understanding

Every Gemini model takes video and audio, not only text and images.

Long documents

1M-token context across the line-up, including the cheapest Flash-Lite.

Google Cloud workloads

Native in Vertex AI, BigQuery and Google Workspace.

Strengths and limits

Strengths

  • Native video and audio input.
  • 1M context everywhere; Gemini 4 Hydrogen can write up to 1M output tokens.
  • Generous free tier in Google AI Studio.

Watch out for

  • Deprecated 2.5 endpoints are rerouted to newer models (and billed at their rates) instead of failing; check outputs after 1 December 2026.
  • Many models stay in preview for months.

About Gemini

Gemini is Google DeepMind’s family of multimodal models, launched in December 2023 to replace PaLM. From the start Gemini was trained on text, images, audio and video together, and every current model still takes all four as input.

Gemini 2.5 (2025) brought “thinking” to the whole line-up, and the 3.x Flash models followed through 2026. The Gemini 4 generation uses element names: Carbon (November 2026) is the strongest reasoning model, Argon (October 2026) is a faster and cheaper frontier model, and Hydrogen (November 2026) is for everyday coding and agent work, with up to 1M output tokens. All of them have a 1M-token context window.

Google handles retirements differently from other vendors. When gemini-2.5-pro and gemini-2.5-flash are shut down on 1 December 2026, requests will be rerouted to newer models (and billed at their prices) instead of failing.

Gemini guides

AI & LLMs Claude Code vs Codex CLI vs Antigravity CLI: Which AI Coding Agent to Use? Updated Nov 2026 AI & LLMs Gemini CLI Is Retired: How to Install Antigravity CLI on Linux Updated Nov 2026

Frequently asked questions

What is the latest Gemini model?

The newest Gemini model is Gemini 4 Hydrogen, released in Nov 2026. It has a 1M-token context window and supports tool calling, reasoning mode, image input, audio input.

Which Gemini models are available in the API?

As of 14 Nov 2026, the current models are Gemini 4 Hydrogen (gemini-4-hydrogen), Gemini 4 Carbon (gemini-4-carbon), Gemini 4 Argon (gemini-4-argon), Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite).

Which Gemini models are being retired?

Announced retirements: Gemini 2.5 Pro on 1 Dec 2026 (replacement: Gemini 3.8 Flash (automatic reroute)); Gemini 2.5 Flash on 1 Dec 2026 (replacement: Gemini 3.5 Flash-Lite (automatic reroute)); Gemini 3.1 Flash-Lite on 7 May 2027 (replacement: Gemini 3.5 Flash-Lite).

Can I download Gemini and run it locally?

No. Gemini models are only available through the vendor's API and partner clouds; the weights are not published.

What happens to Gemini 2.5 Pro after December 1, 2026?

Requests to gemini-2.5-pro are automatically routed to Gemini 3.8 Flash and billed at its rates; gemini-2.5-flash requests go to Gemini 3.5 Flash-Lite. Gemini 2.5 Flash-Lite is not affected.

Compare Gemini

← All AI models

Back to top