Gemini 4 Hydrogen vs Gemini 4 Argon vs Gemini 4 Carbon: API Prices and App Plans

The short version

• Google positions Gemini 4 Hydrogen for fast everyday coding and agents, Gemini 4 Argon for advanced tasks with fewer tokens than Carbon, and Gemini 4 Carbon as its strongest model.
• Hydrogen has the lowest Standard API token prices among these three: US$1.50 for input and US$7.50 for output per million tokens.
• Hydrogen also has rate-limited API free access; Argon and Carbon require the API paid tier.

Which Gemini 4 model should I choose?

Google recommends Gemini 4 Hydrogen for everyday development and agent workflows, Gemini 4 Argon for most demanding long-running tasks, and Gemini 4 Carbon for problems requiring its deepest reasoning. Its Gemini 4 Hydrogen guidance separates the models by reasoning effort, response time and token use.

For Gemini 4 Hydrogen vs Argon, Google’s distinction is everyday execution versus advanced work over longer tasks. For Gemini 4 Argon vs Carbon, Google says Argon responds faster and consumes fewer tokens per turn, while Carbon spends additional thinking tokens exploring possibilities and checking its answers.

How much do Gemini 4 Hydrogen, Gemini 4 Argon and Gemini 4 Carbon cost?

Google lists Standard API input and output rates of US$1.50 and US$7.50 for Hydrogen, US$2.00 and US$10.00 for Argon, and US$2.00 and US$10.00 for Carbon, per million tokens. These are US-dollar API charges from Google’s pages read on November 12, 2026; Gemini app access follows separate Google AI plan requirements.

API prices, token limits and app access

The Gemini 4 models compared share the same total context budget. Longer API responses require background execution or continuation, and thinking consumes part of the available token allowance.

Google’s prices count thinking tokens as output. Google also says Carbon generally uses more thinking tokens per turn than Argon, so their matching token rates do not mean matching charges for a task.

Google bills all processed tokens, including internal instructions for tools and grounding-related function declarations. Search grounding also carries a separate charge for each query performed.

For paid-tier Search grounding, Google lists a monthly allowance of 5,000 free requests shared by Gemini 3 and subsequent models. Further requests cost US$14 per 1,000, with a daily ceiling of 1,500 requests. These US-dollar rates were read on November 12, 2026.

Google lists a 50% reduction in API costs for Batch processing on its pricing page read November 12, 2026.

Which Gemini 4 model is the smartest?

Google calls Gemini 4 Carbon its strongest reasoning model. Google recommends it for demanding architecture decisions, difficult debugging, scientific reasoning and security reviews, with extra thinking devoted to examining possibilities and checking edge cases.

Which Gemini 4 model is the fastest and cheapest?

Google describes Gemini 4 Hydrogen as fast and efficient for coding and agent workflows, and its Standard API prices are the lowest of these three. Hydrogen costs US$1.50 for input and US$7.50 for output per million tokens; Argon and Carbon each cost US$2.00 and US$10.00 through November 30, 2026.

Google separately describes Argon as faster than Carbon, with fewer tokens consumed per turn. Google’s guidance reserves Gemini 3.5 Flash-Lite for workloads prioritizing its lowest cost and shortest response times.

Which Gemini 4 models do I get with a free, Google AI Pro or Google AI Ultra account?

Google’s personal-account rules give free users potential Hydrogen access through Auto, Pro subscribers selectable Hydrogen and Argon, and Ultra subscribers selectable Hydrogen, Argon and Carbon. The Gemini Apps model and plan rules distinguish automatic routing from choosing a model yourself.

• Free account: Auto or Flash-Lite in Chats. With automatic selection enabled, more demanding prompts may route to Hydrogen. Tasks are unavailable.
• Google AI Plus: Hydrogen is selectable in Chats. Tasks are unavailable.
• Google AI Pro: Hydrogen and Argon are available in Chats and Tasks.
• Google AI Ultra: Hydrogen, Argon and Carbon are available in Chats and Tasks.

Our menu records cover desktop Chrome on Windows with a Google AI Pro account on November 12, 2026, and a Pixel 7 Pro using Google app 17.74.14 with a free account on November 13, 2026.

The Pro account showed Everyday help with 4 Hydrogen, Code & analysis with 4 Argon, and a locked Most complex work with 4 Carbon. The free account showed Auto and a locked Code & analysis with 4 Argon.

In the app, Google offers Low, Medium and High thinking levels, with Low as the default when you make no selection. Model access also depends on remaining usage; Google says paid-plan chats continue with Flash-Lite after the applicable limit is reached.

Which Gemini 4 model should I use for coding and agents?

Google recommends Hydrogen for everyday coding and execution, Argon for most advanced engineering tasks, and Carbon for the hardest reasoning problems. Google’s workload guidance is:

• Gemini 4 Hydrogen: everyday code changes, interactive web applications, agents that call tools, document workflows spanning different media, and parallel subagents.
• Gemini 4 Argon: most advanced tasks that run across many steps, with lower token consumption than Carbon.
• Gemini 4 Carbon: difficult architectural work, root-cause debugging, scientific problems and security audits.
• Gemini 3.5 Flash-Lite: high-volume classification and routing, particularly when Google’s lowest-cost option and sub-second response targets matter.

Google also recommends combining Hydrogen and Carbon: let Carbon design interfaces or resolve difficult blockers, then assign edits, parallel subagent work and test execution to Hydrogen. Its API guidance supports changing models between conversation turns while preserving the conversation state.

For Hydrogen’s API thinking settings, Google recommends starting at high for complex work. Google reserves xhigh for difficult multi-step problems because it increases response time and token consumption.

API settings to check when changing models

Hydrogen uses high by default, whereas Gemini 3.8 Flash uses medium. Hydrogen adds xhigh, but a request specifying minimal returns an error.

Hydrogen uses implicit caching rather than explicit cache creation. Remove references to cachedContents and caches.create, along with the request field cached_content. Automatic caching applies once a prompt reaches 4,096 tokens; a cache hit reduces the cached-input charge by 95%.

Remove the sampling fields temperature, top_p and top_k when migrating to Hydrogen. Custom values for those fields are ignored by the server.

For longer responses, setting background to true also requires an explicit max_output_tokens value above 65,536. Without that value, background generation still ends at 65,536 tokens.

For Argon requests with a lower output allowance, Google recommends increasing max_output_tokens to 32,768 or more. Google says Argon can devote more of that allowance to thinking than previous models.

Google says difficult Argon and Carbon requests may need several minutes. Its migration guidance recommends background execution or an HTTP timeout of 15 minutes or more.

How do Gemini 4 Argon and Gemini 4 Carbon prices change on December 1, 2026?

Argon and Carbon move to US$4.00 for input and US$20.00 for output per million tokens on December 1, 2026. Their introductory Standard API prices remain US$2.00 for input and US$10.00 for output through November 30, 2026.

Their Standard cached-input rate also changes, from US$0.10 to US$0.20 per million tokens. Hydrogen’s listed Standard rates remain US$1.50 for input, US$7.50 for output and US$0.075 for cached input per million tokens.

FAQs

Which Gemini 4 model does Google call the smartest?

Google identifies Gemini 4 Carbon as its strongest reasoning model. Google recommends it for especially demanding engineering, science and security tasks.

Which Gemini 4 model does Google describe as fast and efficient?

Google describes Gemini 4 Hydrogen that way for everyday coding and agent workflows. Google also says Argon responds faster than Carbon while consuming fewer tokens per turn.

Which Gemini 4 model has the cheapest API token prices?

Hydrogen has the lowest Standard API rates among these three on Google’s pricing page read November 12, 2026: US$1.50 per million input tokens and US$7.50 per million output tokens. Hydrogen also has a rate-limited free tier.

Is Gemini 4 Hydrogen better than Gemini 4 Argon?

Google assigns them different roles: Hydrogen handles everyday coding and agent execution, while Argon handles most advanced tasks requiring sustained reasoning. Hydrogen has lower Standard API token rates and free-tier access.

Is Gemini 4 Carbon free?

No. Carbon requires the API paid tier, and access through the Gemini app requires Google AI Ultra.

Can I use Gemini 4 Carbon in the Gemini app with Google AI Pro?

No. Google’s personal-account plan rules reserve Carbon for Google AI Ultra. Pro includes Hydrogen and Argon in Chats and Tasks.

Which Gemini 4 model is best for coding?

Google recommends Hydrogen for everyday coding, Argon for most advanced software work, and Carbon for particularly difficult architecture and debugging problems. Google also suggests using Carbon for planning and Hydrogen for edits and test execution.

Which Gemini 4 model is available on the Gemini API free tier?

Hydrogen is available with rate limits and free input and output tokens. Argon and Carbon require paid API access.