Field notes · AI agents, Hermes & Claude Code By Julian Goldie

Gemini 4 Hydrogen vs Argon: Thinking Levels, Limits and Which One to Run

An operator comparison of Google's fast Gemini 4 model and its efficient frontier reasoner.

Get The AI Profit Stack Join AIPB →
  • 01 1,000+ done-for-you AI agent workflows
  • 02 5 live coaching calls / week with me
  • 03 7-day refund, cancel anytime
  • 04 3,400+ AI operators inside

Gemini 4 Hydrogen vs Argon comes down to one trade: Hydrogen is the fast, cheaper workhorse Google released on 12 November 2026, and Argon is the deeper frontier reasoner it shipped on 13 October.

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

Both take the same 1M-token window and the same built-in tools, so for an operator the real differences sit in thinking levels, per-hop output limits, tier access and how many tokens each burns per turn.

If you want a team to wire models like these into your SEO and content work, get your Backlink Gameplan, or try Polar Rank free for 3 days.

Every spec below comes from Google's Gemini API changelog, the two What's new pages and the DeepMind model cards.

Gemini 4 Hydrogen vs Argon at a glance

Specgemini-4-hydrogengemini-4-argon
General availability12 November 202613 October 2026
Google's labelDefault general-purpose modelFlagship reasoning (with Carbon)
Thinking levelslow, medium, high (default), xhighlow, medium (default), high
Context1,048,576 tokens, shared by input, thinking and outputSame unified 1,048,576-token window
Sync output cap65,536 tokens65,536 tokens
Long decodeUp to 1M tokensUp to 1M tokens
Non-streaming REST hopUp to 65,536 tokensUp to 32,768 tokens
TiersFree and paidPaid tier pricing published
LineageAdapted from Gemini 4 CarbonNew model, not a fine-tune

The lineage row matters more than it looks.

The Argon model card says Argon is a new sparse mixture-of-experts model, while the Hydrogen card says Hydrogen is adapted from Carbon, which is itself based on Argon.

So Hydrogen is the distilled, fast branch of the same family, not a cousin from another line.

Thinking levels are the main control

Google's thinking guide treats thinking_level as the dial for quality, latency and spend.

Hydrogen defaults to high and adds a fourth level, xhigh, for hard multi-step coding and agent problems.

Argon defaults to medium and caps at high.

That means the same request with no thinking setting behaves differently on each model: Hydrogen starts higher on its own scale, Argon starts in the middle.

from google import genai
client = genai.Client()

# Hydrogen: drop to medium for interactive loops
fast = client.interactions.create(
    model="gemini-4-hydrogen",
    input="Refactor utils.py into smaller functions",
    generation_config={"thinking_level": "medium"},
)

# Argon: raise to high for the hard planning turn
deep = client.interactions.create(
    model="gemini-4-argon",
    input="Design the migration plan for our billing service",
    generation_config={"thinking_level": "high"},
)

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

Google warns that xhigh on Hydrogen raises latency and token use and can overthink simple prompts.

Run your eval set at every level, then keep the lowest one that clears your quality bar.

Output limits and long decode

Both models share one 1,048,576-token budget across input, thinking and output, per the long decode guide.

A single synchronous response stops at 65,536 tokens on either model.

To go further, set background to true and raise max_output_tokens above 65,536, and Google's servers continue in 64K hops.

The one operator gotcha: Hydrogen allows up to 65,536 tokens per non-streaming hop, while Carbon and Argon allow 32,768 per non-streaming REST hop.

If you call Argon over plain REST without streaming, plan for the smaller hop.

🔥 Running Gemini 4 inside a real agent stack? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members testing the newest models on real business work. → Join the AI Profit Boardroom

Published scores from Google's own model cards

Google does not publish one table with both models side by side, so the rows below combine the Hydrogen card (November 2026) and the Carbon card, which lists Argon (October 2026).

Benchmark (Google-reported)HydrogenArgonCarbon
SWE-sweep (agentic coding)13.5%4.3%9.6%
OSWorld-2.1, offline subset (computer use)77.6%74.7%81.5%
Chartography (charts and figures)66.6%71.6%72.5%

Read it plainly: Hydrogen posts the higher SWE-sweep and OSWorld numbers in these cards, and Argon reads charts better.

The Carbon card also shows Argon behind Carbon on most reasoning-heavy rows, which fits Google's positioning.

Treat vendor numbers as a starting point and test on your own repository before switching.

Where Hydrogen wins for operators

  • Everyday coding, interactive web apps and tool-using agents, which is the job Google gives it.
  • Parallel subagents: Google says it batches read and search tool calls in parallel and adapts better when tool calls fail.
  • Upgrades from Gemini 3.8 Flash: the Hydrogen migration notes call it the recommended upgrade.
  • Free-tier testing, because Hydrogen is listed on both the free and paid tiers.

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

Where Argon wins

  • Long-horizon tasks where you want frontier reasoning but lower token use than Carbon.
  • Multi-turn deep investigations, legal and financial analysis and dense document synthesis, the uses its model card lists.
  • Workloads where chart and figure reading drives the answer.
  • Teams already tuned to medium as the default thinking level.

📞 Want Gemini 4 working on your search traffic? Get your Backlink Gameplan and we will map where fast models and deep models fit into your SEO and AI search work, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days

Migration notes from Gemini 3.8 Flash to Hydrogen

  1. Change the model string to gemini-4-hydrogen.
  2. Remove hardcoded low max_output_tokens caps such as 8,192, or raise them to 65,536.
  3. Stop sending minimal as a thinking level; Hydrogen rejects it with an error.
  4. Remove explicit caching calls; implicit caching runs on prompts of at least 4,096 tokens.
  5. Remove temperature, top_p and top_k; out-of-range values return a 400.
  6. Declare tools natively instead of asking for raw-text JSON tool calls.

Pairing both models in one session

Google documents a split you can copy directly: plan on a frontier model, execute on Hydrogen.

In the Interactions API you can route a planning turn to Argon or Carbon and the execution turns to Hydrogen inside one multi-turn session by passing previous_interaction_id.

That keeps the expensive thinking where it pays off and moves the volume to the cheaper model.

Cost per turn, without the price list

Google's latest models page lists Hydrogen at a lower per-token price than Argon.

Argon is on an introductory rate through 30 November 2026, and its listed standard rate after that is double the introductory one.

Per-token price is only half the story, because thinking tokens are billed too, and Hydrogen at xhigh can spend far more than Argon at medium on the same prompt.

Measure tokens per finished task, not tokens per request.

Common mistakes

  • Comparing Hydrogen at its high default with Argon at its medium default and calling it a fair test.
  • Leaving an old 8K output cap in place, then blaming the model for cut-off answers.
  • Forgetting that thinking and output share one budget, so a tight max_output_tokens starves the answer.
  • Assuming Argon is on the free tier because Hydrogen is.

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

What Argon's own card claims

Argon's model card compares it with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 rather than with Hydrogen.

On that card Argon posts 77.9% on DeepSWE v1.1, 91.9% on Vibe Code Bench and 57.4% on Terminal-bench 4.0.

Its long-context rows are the standout: 99.7% on GraphWalks up to 128K tokens and 84.2% from 256K to 1M tokens.

If your agent reads whole codebases or long transcripts in one go, that long-context row is the one to weigh.

Long decode in practice

Long decode is where both models break past the 64K wall, and the call shape is identical.

Set background=True, raise max_output_tokens, then poll until the interaction finishes.

import time
from google import genai
client = genai.Client()

job = client.interactions.create(
    model="gemini-4-hydrogen",          # or "gemini-4-argon"
    input="Write the full test suite for the payments module.",
    background=True,
    generation_config={"max_output_tokens": 262144,
                       "thinking_level": "high"},
)
while job.status in ("in_progress", "queued"):
    time.sleep(5)
    job = client.interactions.get(id=job.id)
# each 64K hop adds a model_output step: join them all

Google's long decode page says each hop adds a model_output step to interaction.steps, so join the text across every step when the job ends.

Forget that and you will ship only the first 64K tokens of a long answer.

Caching and token billing differences

Neither model supports explicit caching, per the caching docs, so cachedContents and caches.create are out.

Hydrogen's guide says implicit caching runs by default on prompts of at least 4,096 tokens and discounts cached input tokens by 95% on a hit.

Google also bills every token the model processes, including thinking tokens and the internal instructions it adds for grounding and function declarations.

For agent loops that resend a big system prompt every turn, implicit caching is the biggest lever you have on cost.

Running either model inside an agent framework

Hermes Agent lists gemini-4-hydrogen, gemini-4-carbon and gemini-4-argon in its Google Gemini guide, and maps its own reasoning effort onto Google's thinkingLevel.

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

On Hydrogen, Hermes sends xhigh for its xhigh, max and ultra efforts; on Argon those efforts all map to high with a bigger long-decode budget.

Google's own Antigravity agent in Gemini Managed Agents now runs Hydrogen by default, and Google's Argon guide says the same agent supports Argon too.

Thought signatures and multi-turn state

One migration detail bites agent builders: Google asks you to keep conversation state intact across turns.

In the Interactions API that means passing previous_interaction_id; in a client-managed generateContent history it means returning prior model turns and thought signatures unchanged.

Google also says not to copy one thought signature across parallel functionCall parts, which is an easy bug to write when you fan out tool calls.

Picking a thinking level per agent role

In a multi-agent setup, give each role its own level rather than one global setting.

A router or classifier step rarely needs more than low on Hydrogen.

Workers that edit files and run tests do well on Hydrogen at medium or high.

A planner or reviewer that must catch subtle mistakes is the place for Argon at high, or Hydrogen at xhigh if you want to stay on one model.

Log tokens per role for a week and the right split usually becomes obvious.

Which one to pick this week

Default to Hydrogen for build-and-iterate work, agents that call lots of tools, and anything you run at volume.

Move a task to Argon when Hydrogen at high keeps missing, or when you need deeper reasoning with less spend than Carbon.

For agent workflows built on models like these, the AI Profit Boardroom has the trainings and the Agent OS, and you can get your Backlink Gameplan if you want the SEO side handled, or try Polar Rank free for 3 days.

Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →

FAQ: Gemini 4 Hydrogen vs Argon

Is Gemini 4 Hydrogen better than Argon?

It depends on the job. Google positions Hydrogen for fast everyday coding and agents and Argon for deeper frontier reasoning; on Google's own cards Hydrogen scores higher on SWE-sweep and OSWorld-2.1 while Argon is ahead on Chartography.

What is the default thinking level on each model?

Hydrogen defaults to high and adds xhigh. Argon defaults to medium and its highest level is high.

Do Hydrogen and Argon have the same context window?

Yes. Both use a unified 1,048,576-token window shared by input, thinking and output, with long decode up to 1M tokens.

Is Gemini 4 Hydrogen free?

Google lists Hydrogen on both the free and paid tiers of the Gemini API and Google AI Studio.

When were they released?

Argon became generally available on 13 October 2026 and Hydrogen on 12 November 2026, per the Gemini API changelog.

Does Argon support long decode?

Yes. Both Argon and Hydrogen can generate past 64K tokens, up to the 1M unified window, using background execution.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

  • I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
  • I've grown a YouTube channel to 400,000+ subscribers.
  • I wrote "Link Building Mastery".
  • My Udemy courses have taught over 29,000 students.

→ Get my best AI training inside the AI Profit Boardroom

Run one task on each with matched thinking levels and you will settle Gemini 4 Hydrogen vs Argon for your own stack; to put the winner to work on SEO, get your Backlink Gameplan, or try Polar Rank free for 3 days.

§ Member reviews

What members say about the AI Profit Boardroom

Read the member reviews on Skool →
Ahmad Bukhari: Julian doesn't gatekeep. That's why I'd recommend this group. He's extremely active here, sharing what's working right now and jumping in to help anyone who's stuck. If he can give it away, he does. He keeps the vibe positive and pushes everyone to help each other. His challenges reward members for getting people unstuck, so the whole room shares knowledge, not just him. Want a generous, positive place to learn AI and actually use it? This is it. Thanks Julian!
Ahmad Bukhari
Julian doesn't gatekeep. That's why I'd recommend this group. He's extremely active here, sharing what's working right now and jumping in to help anyone who's stuck. If he can give it away, he does. He keeps the vibe positive and pushes everyone to help each other. His challenges reward members for getting people unstuck, so the whole room shares knowledge, not just him. Want a generous, positive place to learn AI and actually use it? This is it. Thanks Julian!

Still a paying member after 1 month

Jay Parsons: There are so many helpful members in this amazing community ready to help out with suggestions and solutions whenever someone asks a question. I am amazed at the progress I have made in learning AI. I came here to learn how to use AI better and apply what I learned to my business. I am learning so much more than I even dreamed of. I have built an Agentic OS and use it every day, and I have no background in coding. There are an incredible amount of useful tutorials on everything AI.
Jay Parsons
There are so many helpful members in this amazing community ready to help out with suggestions and solutions whenever someone asks a question. I am amazed at the progress I have made in learning AI. I came here to learn how to use AI better and apply what I learned to my business. I am learning so much more than I even dreamed of. I have built an Agentic OS and use it every day, and I have no background in coding. There are an incredible amount of useful tutorials on everything AI.

Still a paying member after 11 months

Brian O'Connor: At 47, with 30 years in sales and finance, embracing AI was exciting but overwhelming. While building a business to help others adopt AI, I found myself chasing tools and fearing I was falling behind. Julian and the AI Profit Boardroom helped me refocus through practical, personalized advice and a supportive community. I now feel more confident adapting to change while staying focused on my goals. Thank you, Julian, for bringing this community together!
Brian O'Connor
At 47, with 30 years in sales and finance, embracing AI was exciting but overwhelming. While building a business to help others adopt AI, I found myself chasing tools and fearing I was falling behind. Julian and the AI Profit Boardroom helped me refocus through practical, personalized advice and a supportive community. I now feel more confident adapting to change while staying focused on my goals. Thank you, Julian, for bringing this community together!

Still a paying member after 2 months

Neil Martin: This is a truly amazing community, thanks in no small part to Julian's prodigious output and tireless efforts to provide value to his subscribers. Nothing is too much trouble for him. If you ask a question it gets answered in short order usually with a custom video. I have been in quite a few groups and this level of service is unparalleled. If you are serious about getting an AI education, you need to be in this group!
Neil Martin
This is a truly amazing community, thanks in no small part to Julian's prodigious output and tireless efforts to provide value to his subscribers. Nothing is too much trouble for him. If you ask a question it gets answered in short order usually with a custom video. I have been in quite a few groups and this level of service is unparalleled. If you are serious about getting an AI education, you need to be in this group!

Still a paying member after 7 months

Crystal Hill: AIPB has inspired me to go beyond learning AI as individual tools, skills, and prompts, and start thinking in terms of larger systems. What I really appreciate is how responsive Julian is and how quickly the material evolves with the technology. You’re being exposed to the right tools and ideas as they emerge. AIPB has given me the confidence to learn, experiment, build, and think much bigger about what’s really possible with AI.
Crystal Hill
AIPB has inspired me to go beyond learning AI as individual tools, skills, and prompts, and start thinking in terms of larger systems. What I really appreciate is how responsive Julian is and how quickly the material evolves with the technology. You’re being exposed to the right tools and ideas as they emerge. AIPB has given me the confidence to learn, experiment, build, and think much bigger about what’s really possible with AI.

Still a paying member after 5 months

Maurice Morrison: AI Profit Board Room has been a game-changer. The insights are practical, timely, and focused on turning AI into real business results—not just theory or hype. The community is supportive, the guidance is clear, and I leave every session with ideas I can actually apply. If you’re serious about using AI to grow smarter and move faster, this is absolutely worth joining.
Maurice Morrison
AI Profit Board Room has been a game-changer. The insights are practical, timely, and focused on turning AI into real business results—not just theory or hype. The community is supportive, the guidance is clear, and I leave every session with ideas I can actually apply. If you’re serious about using AI to grow smarter and move faster, this is absolutely worth joining.

Still a paying member after 2 months

Jeremy Murdock: I have truly enjoyed my experience here. The group of folks on the calls is quite diverse and I find it comforting to know that I am not the only one going through the changes that are happening at a faster and faster pace. I have chosen to embrace the change and go full force into integrating AI and what I have learned form the group and Julian. I am also impressed and appreciative that he actually takes the time to help with challenges I have along the way and get involved personally!
Jeremy Murdock
I have truly enjoyed my experience here. The group of folks on the calls is quite diverse and I find it comforting to know that I am not the only one going through the changes that are happening at a faster and faster pace. I have chosen to embrace the change and go full force into integrating AI and what I have learned form the group and Julian. I am also impressed and appreciative that he actually takes the time to help with challenges I have along the way and get involved personally!

Still a paying member after 10 months

Michael Lovitch: I have been a part of a million ai groups in Skool for years now. They are all okay and have helped, but usually they get out of date and don't offer real service to help you through especially at the beginning. This group is different, Julian actually answers questions which is crazy! And although ai marketing moves at the speed of light, they stay current and actually help you keep up… Highly recommended!
Michael Lovitch
I have been a part of a million ai groups in Skool for years now. They are all okay and have helped, but usually they get out of date and don't offer real service to help you through especially at the beginning. This group is different, Julian actually answers questions which is crazy! And although ai marketing moves at the speed of light, they stay current and actually help you keep up… Highly recommended!

Still a paying member after 3 months

Hoebing Lo: I don't know of any other community that is as active and sharing. Julian is a star and he answers so many questions in lightning speed. If you want a great place to start with AI here it is. No question too simple. No workflow too complicated. The community is about action and results not education so if you are looking to win at AI then come for the ride of a lifetime!
Hoebing Lo
I don't know of any other community that is as active and sharing. Julian is a star and he answers so many questions in lightning speed. If you want a great place to start with AI here it is. No question too simple. No workflow too complicated. The community is about action and results not education so if you are looking to win at AI then come for the ride of a lifetime!

Still a paying member after 5 months

Darren Wagstaff: Very responsive... great personal support (not an ai chat bot). Very clear straight forward trainings in easy to understand actionable language. I'm in numerous Skool groups but this one has the most human and real community by far. No question gets ignored. You never fell like just another number. It's great to see so many members getting great results with Ai.
Darren Wagstaff
Very responsive... great personal support (not an ai chat bot). Very clear straight forward trainings in easy to understand actionable language. I'm in numerous Skool groups but this one has the most human and real community by far. No question gets ignored. You never fell like just another number. It's great to see so many members getting great results with Ai.

Still a paying member after 2 months

Rod Martin: Julian is brilliant, and more prolific than I can keep up with (which is a great thing). His AgentOS is remarkable, and he is THE go-to guy for what you can really do (and not do) with the ever-increasing deluge of new AI upgrades and tools. HIGHLY recommend.
Rod Martin
Julian is brilliant, and more prolific than I can keep up with (which is a great thing). His AgentOS is remarkable, and he is THE go-to guy for what you can really do (and not do) with the ever-increasing deluge of new AI upgrades and tools. HIGHLY recommend.

Still a paying member after 5 months

Yosuke Sato: The best Skool course you get for AI! I have attended a few Skool courses before, but AI Profit Boardroom is the most active community and has great support from the management team, including Julian! You can get the most up-to-date information and practical implementations in this community. I highly recommend this!
Yosuke Sato
The best Skool course you get for AI! I have attended a few Skool courses before, but AI Profit Boardroom is the most active community and has great support from the management team, including Julian! You can get the most up-to-date information and practical implementations in this community. I highly recommend this!

Still a paying member after 19 months

⚑ Next step

Ready to run AI agents that actually do the work?

Join 3,400+ operators inside the AI Profit Boardroom. Get 1,000+ plug-and-play AI agent workflows, 5 coaching calls a week, and a community that holds you accountable.

Join The AI Agent Community →

7-day no-questions refund · Cancel anytime

Found this useful? Get more of this site in your Google.

One tap adds us as a preferred source — Google then shows you our posts more often in Search, Top Stories, and AI Overviews.

← Back to all posts