Key takeaways

  • As of November 15, 2026, Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, against $2 and $10 for Argon, which rises to $4 and $20 on December 1.
  • Artificial Analysis scores Hydrogen 51 on high and 54 on xhigh against 50 for Argon, and measures it at 170 output tokens per second against 65.
  • Hydrogen is on the Gemini API free tier and reaches free Gemini app users, while Argon needs a paid API tier or a Google AI Pro or Ultra plan.

Gemini 4 Hydrogen vs Argon comes down to this: Hydrogen scores about the same as Argon on independent tests, runs more than twice as fast and costs less. As of November 15, 2026, Hydrogen is $1.50 per million input tokens and $7.50 per million output tokens on the Gemini API pricing page, against $2 and $10 for Argon, and Argon's price doubles to $4 and $20 on December 1.

Argon still has a place. Google built it for deep, multi-step reasoning in science, finance, law and security, and it has published results there that Hydrogen hasn't matched or tried. But for most coding and agent work, Hydrogen is now the better default. This comparison is based on Google's docs, Google's announcements and Artificial Analysis's published results, not on our own tests.

The two models in one paragraph each

Gemini 4 Argon came out on October 13, 2026. Google's model docs call it a more efficient frontier reasoning model, meaning it answers faster and uses fewer tokens per turn than Gemini 4 Carbon, the strongest model in the family. Google aimed it at coding, finance, legal work, cybersecurity and science.

Gemini 4 Hydrogen came out on November 12, 2026. Google calls it "our daily driver for agentic workflows" and pitches it at fast, everyday coding and agent tasks, leaving long-horizon reasoning to Carbon and Argon. Our news story on the Hydrogen launch has the full announcement.

Gemini 4 Hydrogen vs Argon at a glance

Gemini 4 Hydrogen Gemini 4 Argon
Released November 12, 2026 October 13, 2026
API model code gemini-4-hydrogen gemini-4-argon
Input price (per 1M tokens, as of November 15, 2026) $1.50 $2.00 until November 30, then $4.00
Output price (per 1M tokens, as of November 15, 2026) $7.50 $10.00 until November 30, then $20.00
Cached input (per 1M tokens) $0.075 $0.10, then $0.20
Batch and Flex (input / output) $0.75 / $3.75 $1.00 / $5.00, then $2.00 / $10.00
API free tier Yes, on Standard No
Context window 1M tokens 1M tokens
Max output (Google's models page) 64K / 1M tokens 64K / 1M tokens
Inputs Text, image, audio, video, PDF Text, image, audio, video, PDF
Artificial Analysis Intelligence Index 51 (high), 54 (xhigh) 50 (high), 48 (medium)
Output speed, Artificial Analysis 170 tokens/s (high) 65 tokens/s (high)
Time to first answer token 16.95 s (high) 43.26 s (high)
Gemini app Free users (auto model selection) and paid plans Google AI Pro and Ultra

How much Hydrogen and Argon cost

As of November 15, 2026, the Gemini API pricing page lists Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens on the Standard tier, with no higher rate for long prompts. Batch and Flex, the two slower and cheaper ways to run requests, are half that: $0.75 and $3.75. Priority, for traffic that must not be delayed, is $2.70 and $13.50.

Argon is $2 and $10 on Standard through November 30, 2026. From December 1 the page lists $4 and $20. Batch and Flex go from $1 and $5 to $2 and $10, and Priority from $3.60 and $18 to $7.20 and $36. (Google's Argon announcement says the increase starts "from November 30", while the pricing page says the lower price runs through November 30. The pricing page is the one that bills you.)

A concrete job makes the gap clearer. Sending 1 million input tokens and getting 200,000 tokens back costs $3.00 on Hydrogen. On Argon it costs $4.00 today and $8.00 from December 1. Argon's December price is 2.67 times Hydrogen's on both input and output.

Both models get cached input at 95% off, and both bill thinking tokens as output. Grounding with Google Search costs the same for both on paid plans: 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000.

Context window and output length

Both models read up to 1 million tokens, accept the same inputs (text, images, audio, video and PDFs), and list the same "64K / 1M tokens" output limit on Google's models page. Google's Hydrogen post says it can write up to 1 million tokens, and Argon's post describes a 1M token limit shared between input and output. On context, there is no reason to pick one over the other.

Speed: Hydrogen is about 2.6 times faster

Artificial Analysis measured Hydrogen at 170 output tokens per second at its default high setting, and between 169 and 175 at every other setting. It measured Argon at 65 tokens per second on high and 56 on medium.

Waiting time follows the same pattern. On high, Hydrogen took 16.95 seconds to start its answer and Argon took 43.26 seconds, because both think first. Hydrogen's low setting starts in 5.67 seconds, which suits chat and quick tool calls.

Google's own speed claim compares Hydrogen with Gemini 3.8 Flash, not Argon: 2.5 times faster at the default setting.

Benchmarks: a tie on the shared index, different strengths elsewhere

The one score both models share is the Artificial Analysis Intelligence Index, which combines many tests into one number. Hydrogen scores 51 on high and 54 on xhigh, a new top thinking level. Argon scores 50 on high. Google quotes the same figures in its Hydrogen post.

Hydrogen also used fewer resources to get there. Artificial Analysis lists a cost of $3.16 per index task for Hydrogen on xhigh and $4.97 for Argon, at Argon's current introductory price. The trade-off is tokens: Hydrogen on xhigh used 304 million output tokens on the index, against 110 million on high and 113 million for Argon on high.

Beyond that, Google tested the two models on different things, so their numbers don't line up:

  • Hydrogen (Google's figures): 57% on Terminal-Bench 4.1, a test of agents working in a terminal, against 18.2% for Gemini 3.8 Flash. 58.8% on HLE-Diamond, 52.3% on Vals Tax Agent Bench (ranked first), 55.8% on Vals Legal Research Bench (tied for first) and 71.6% on Agents' Last Exam v1.1.
  • Argon (Google's figures): 87% on ERQA, a test of reasoning about the physical world, 57% on RareBench for rare-disease diagnosis, 52.2% on LabBench for lab reasoning, and a 0.7% attack success rate on Gray Swan's prompt injection test. Google says Argon ranks first on the Vals Index.

Treat the company figures as claims. Independent numbers so far cover the Artificial Analysis index, speed and price, and those favor Hydrogen.

Availability in the Gemini app and the API

In the Gemini app, Hydrogen reaches the most people. Google says free users get it through automatic model selection when it fits the prompt, and the Gemini plans page lists it for Free, Google AI Plus ($4.99 a month), Pro ($19.99) and Ultra (from $99.99). It also runs in Google Search for everyone, in Gemini in Google Sheets for Pro and Ultra subscribers, and in Google Antigravity for paid Pro and Ultra subscribers.

Argon is for paying app users. The Gemini app release notes of October 13 say it rolled out to Google AI Pro and Ultra subscribers, and Google's announcement adds AI Mode in Search for the same plans and Antigravity for Ultra.

In the API, Hydrogen is open to every tier in AI Studio, including the free one. Argon is paid only: the pricing page says "Gemini 4 Argon is not available on the Free tier." Both are in Android Studio and Gemini for business. Neither announcement names Vertex AI.

Which to use for what

Use Hydrogen for coding agents, tool calls, data work, chat and anything run at volume. It matches Argon on the shared index, answers faster, and from December 1 costs well under half as much.

Use Argon when the work is close to what Google tested it on, such as science, medical research or visual and spatial reasoning, and you've checked on your own prompts that it does better. Before December 1, the price gap is small enough to test both.

For the hardest problems, Google points to Carbon. Its Antigravity Orchestrator pairs Carbon with Hydrogen workers, and Google says that mix came close to Carbon's success rate on Terminal-Bench 4.0 with 24% less quota used. Our Gemini 4 Carbon vs Argon comparison covers the top of the family.

Frequently asked questions

Is Gemini 4 Hydrogen free?

Yes, with limits. Free Gemini app users get Hydrogen through automatic model selection, and the Gemini API free tier covers it on the Standard tier. Argon is not on the API free tier and needs Google AI Pro or Ultra in the app.

Gemini 4 Hydrogen vs Carbon: which is better?

Carbon is the stronger model. Artificial Analysis scores it 60 on high against 51 for Hydrogen, but measures it at 65 output tokens per second against Hydrogen's 170. Carbon has the same API price as Argon: $2 and $10 per million tokens until November 30, 2026, then $4 and $20.

Gemini 4 Hydrogen vs Claude Sonnet 5.5: which is faster and cheaper?

Hydrogen, on published numbers. Artificial Analysis measures it at 170 output tokens per second against 142 for Sonnet 5.5, and scores it 54 on xhigh against 50 for Sonnet 5.5 on xhigh and 55 on max. Sonnet 5.5 costs $2 and $10 per million tokens on Anthropic's pricing page, against $1.50 and $7.50 for Hydrogen as of November 15, 2026.

Is Gemini 4 Hydrogen better than Gemini 3.8 Flash?

Google says so. It reports Hydrogen as 2.5 times faster than Gemini 3.8 Flash at the default setting, with 57% on Terminal-Bench 4.1 against 18.2% for 3.8 Flash. Those are Google's own figures.

Sources

  1. Introducing Gemini 4 HydrogenGoogle
  2. Introducing Gemini 4 ArgonGoogle
  3. Gemini 4 Hydrogen: everyday coding and agentic workflows for developersGoogle
  4. Gemini Developer API pricingGoogle
  5. Gemini modelsGoogle
  6. Gemini app release notesGoogle
  7. Google AI plans and subscriptionsGoogle
  8. Gemini 4 Hydrogen: intelligence, performance and price analysisArtificial Analysis
  9. Gemini 4 Argon: intelligence, performance and price analysisArtificial Analysis
  10. Gemini 4 Carbon: intelligence, performance and price analysisArtificial Analysis
  11. Claude Sonnet 5.5: intelligence, performance and price analysisArtificial Analysis
  12. Claude API pricingAnthropic

Add ClaudeGPT Si as a preferred source on Google