Gemini 4
Gemini 4 is Google DeepMind’s model family for coding, analysis, and everyday tasks. Choose Carbon for top-tier reasoning and complex long-horizon software engineering. Select Neon for multi-step coding and analysis at a lower token cost than post-promotion Carbon. Use Hydrogen for fast, affordable everyday coding and chat. Google describes Argon as a faster frontier option than Carbon that uses fewer tokens per turn.
Explore the Gemini API model catalog ↗At a glance
Gemini Developer API. Prices in US dollars per million input and output tokens on Standard. App plans have separate access rules.
| Decision | Developer API availability | Standard input / output | Personal Gemini app |
|---|---|---|---|
| Hydrogen | Free tier and paid | USD 1.50 / 7.50 | Free Auto may choose Hydrogen. Plus and higher can select it. |
| Neon | Paid | USD 3 / 15 | Pro and Ultra |
| Carbon | Paid | USD 2 / 10 through Nov 30. USD 4 / 20 from Dec 1, 2026. | Ultra |
| Argon | Paid | USD 2 / 10 through Nov 30. USD 4 / 20 from Dec 1, 2026. | Absent from the current personal app model table. |
Hydrogen
- Developer API availability
- Free tier and paid
- Standard input / output
- USD 1.50 / 7.50
- Personal Gemini app
- Free Auto may choose Hydrogen. Plus and higher can select it.
Neon
- Developer API availability
- Paid
- Standard input / output
- USD 3 / 15
- Personal Gemini app
- Pro and Ultra
Carbon
- Developer API availability
- Paid
- Standard input / output
- USD 2 / 10 through Nov 30. USD 4 / 20 from Dec 1, 2026.
- Personal Gemini app
- Ultra
Argon
- Developer API availability
- Paid
- Standard input / output
- USD 2 / 10 through Nov 30. USD 4 / 20 from Dec 1, 2026.
- Personal Gemini app
- Absent from the current personal app model table.
Why it matters
Pricing and access vary across developer endpoints and consumer app subscriptions, allowing teams to align speed, reasoning depth, and operational costs.
Which Model to Choose by Job
Select your Gemini 4 model based on project complexity and latency requirements. Deploy Carbon when tackling multi-repository refactoring, complex bug localization, and other hard engineering tasks. Turn to Neon for multi-step agent pipelines, algorithmic data analysis, and pull request reviews where you need deeper reasoning than Hydrogen without Carbon's post-promotion price point. Run Hydrogen for high-throughput tasks such as basic script writing and everyday chat. Google describes Argon as faster than Carbon with fewer tokens per turn. Compare cost, latency and successful outcomes on your own workload.
- Gemini 4 Hydrogen API reference
Check the request model ID, Standard rates, thinking levels and output limits.
- Gemini 4 Neon API reference
Check the request model ID, Standard rates, thinking levels and output limits.
- Gemini 4 Carbon API reference
Check the request model ID, Standard rates, thinking levels and output limits.
- Gemini 4 Argon API reference
Check the request model ID, Standard rates, thinking levels and output limits.
Personal App versus Developer API Access
Google structures model access differently across consumer personal apps and the Developer API. In the personal app, Free tier chats use Auto routing, which can direct requests to Hydrogen or a lightweight model. Free Tasks list Hydrogen. The Plus plan allows manual selection of Hydrogen. The Pro plan adds Neon, while the Ultra plan adds Carbon. Google rolled out Argon to Pro and Ultra users in October 2026, and the model is absent in the current personal app help model picker table. All four models are generally available in the Developer API. Workspace and Google Antigravity have separate access rules.
- Current personal Gemini app model access
Check model choices for a personal Google account.
- Cloud model versions
Check model availability and lifecycle for Google Cloud.
- Google Antigravity models
See model choices and separate eligibility for Antigravity.
Standard USD Developer API Pricing
Developer API pricing is billed in USD per million tokens, with thinking tokens billed as standard output tokens. Carbon and Argon cost $2 per million input tokens and $10 per million output tokens under promotional pricing through November 30, 2026, after which their standard rate becomes $4 per million input tokens and $20 per million output tokens on December 1, 2026. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, alongside an available free API tier. Neon costs $3 per million input tokens and $15 per million output tokens. External tools, connectors, and runtime environments may incur separate charges.
- Standard Gemini Developer API pricing
Check current rates, cache pricing and separate service tiers.
Developer API Limits and Thinking Budgets
All four Gemini 4 models accept multimodal inputs, including text, images, audio, video, and PDF documents, and all four return text outputs exclusively. Each model shares a unified context window of 1,048,576 tokens, a default synchronous output maximum of 65,536 tokens, and an optional long decode mode that supports up to 1,048,576 output tokens within the shared context boundary. Developer API thinking parameters allow developers to manage reasoning depth. Hydrogen and Neon support low, medium, high, and xhigh, with high as the default. Carbon and Argon support low, medium, and high, with medium as the default. In the Interactions API, one non-streaming hop on Carbon or Argon is capped at 32,768 tokens. Streaming hops and Neon or Hydrogen non-streaming hops can use up to 65,536. Long decode can continue within the shared 1,048,576-token context.
- Gemini Developer API model catalog
Read the current model and feature contracts.
- Long decode documentation
Review the request settings and continuation behavior.
Google's Neon MCP and Blender Demonstration
Google demonstrated Gemini 4 Neon's multi-step execution capabilities by connecting the model to Antigravity parallel agents and the external 3D software Blender through the Model Context Protocol (MCP) to render a 3D laboratory bench.
- Google’s Neon launch examples
See how Google used Neon with Antigravity and Blender.
Useful first jobs
- Draft unit tests and check everyday code changes.
- Test a multi-step workflow with a fixed set of tools and check the finished output.
- Refactoring complex enterprise software systems and resolving cross-repository dependencies
Frequently asked questions
Can you access Gemini 4 models for free?
Yes. Google provides free access to Gemini 4 Hydrogen through the Developer API free tier and via the personal consumer app, where the Free plan uses Auto routing that can route chats to Hydrogen. Carbon, Neon, and Argon require paid Developer API accounts.
Which Gemini 4 model should I try first?
Choose Carbon for top-tier reasoning and complex long-horizon software engineering. Select Neon for multi-step coding and analysis at a lower token cost than post-promotion Carbon. Use Hydrogen for fast, affordable everyday coding and chat.
Do Gemini 4 models generate images or 3D files natively?
No. All four Gemini 4 models output text only. While they ingest text, images, audio, video, and PDF files, a visual or 3D artifact can come from external software the model controls, such as driving Blender via the Model Context Protocol.
How do Carbon and Neon compare for development work?
Carbon serves as Google's premier reasoning model for the hardest, long-horizon software engineering problems, priced at $2/$10 per million tokens through November 30, 2026, and $4/$20 starting December 1. Neon provides faster responses and deeper reasoning than Hydrogen at $3 input and $15 output per million tokens. Carbon has lower Standard token rates during the promotion. From December 1, Neon has the lower rates.
Sources and evidence
Checked November 20, 2026 using Google’s dated API changelog, model specifications, Standard pricing and the current personal Gemini app help.
- DeepMind Gemini 4 family ↗
Google · Primary source
- Gemini Developer API changelog ↗
Google · Primary source
- Gemini Developer API model catalog ↗
Google · Primary source
- Standard Developer API pricing ↗
Google · Primary source
- Personal Gemini app models and tiers ↗
Google · Primary source
- Gemini 4 Neon launch examples ↗
Google · Primary source