Google · Released, not on our provider

Gemini 4 models (2026): Carbon, Neon, Hydrogen and Argon compared, with Google’s dates and prices

Google lists four Gemini 4 models today, Argon (October 13, 2026), Carbon (November 10), Hydrogen (November 12) and Neon (November 18), all with a 1M-token context window and priced at $1.50 to $3 per million input tokens, with Hydrogen the cheapest; each of the four runs in the chat on this site.

Released, not on our providerLast updated:

Use it now

Chat

This chat starts with Gemini 4 Hydrogen, not Gemini 4 models.

≈ 2 credits per reply · See all models

New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.

Start chatting — 5 free credits
Maker
Google
Type
Text
Released
No date given
OpenRouter id
None

Checked

Gemini 4 models: facts, with sources

  • Google’s Gemini 4 Hydrogen model card lists results as of November 2026 for Hydrogen against Claude Sonnet 5.5 and GPT-6.1 Terra: SWE-sweep 13.5% against 3.1% and 5.8%, HLE-Diamond 58.8% against 53.3% and 47.3%, and Terminal-Bench 4.1 (cheating-adjusted) 57.0% against 63.9% for Claude Sonnet 5.5 with no figure for Terra. These are Google’s own figures; we did not run them.Source: deepmind.google · checked 2026-11-21
  • The Neon post says Neon scores 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, #3 overall and just behind larger, slower models, and ranks #2 on the Vals Index at 70.9%. These are Google’s claims in its launch post; we did not test them.Source: blog.google · checked 2026-11-21

Our table

Gemini 4 models side by side: dates, prices, thinking levels and where to run them

  • Google’s Gemini API release notes list these general-availability (GA) entries for the Gemini 4 models: Gemini 4 Argon on October 13, 2026, Gemini 4 Carbon on November 10, Gemini 4 Hydrogen on November 12 and Gemini 4 Neon on November 18, 2026, which is the newest entry.Source: aistudio.google.com · checked 2026-11-21
  • Google’s Gemini API pricing page lists these standard paid-tier prices per million input / output tokens: Gemini 4 Hydrogen $1.50 / $7.50; Gemini 4 Neon $3 / $15; Gemini 4 Carbon and Gemini 4 Argon $2 / $10 through November 30, 2026 and $4 / $20 from December 1, 2026. Hydrogen is free of charge on the free tier; Carbon, Argon and Neon are not available on it.Source: aistudio.google.com · checked 2026-11-21
  • Google’s Gemini API model list shows the same specifications on all four Gemini 4 model cards: a 1M-token context window, a maximum output of 64K tokens (up to 1M with long decode) and text, image, audio, video and PDF as the modalities. The model pages give the context window as 1,048,576 tokens.Source: aistudio.google.com · checked 2026-11-21
  • Google’s "Choosing a model" page compares Carbon, Neon and Hydrogen. Its Speed column reads "Fast" for Hydrogen, which it recommends for everyday coding and agent work, "Moderate" for Neon, for complex, interactive software engineering and analysis where someone is waiting on the answer, and "Deliberate" for Carbon, for long-horizon background work such as multi-hour runs, deep research and security audits. Argon is not in that table.Source: aistudio.google.com · checked 2026-11-21
  • The same page lists the thinking levels: Carbon low, medium (the default) and high; Neon low, medium, high (the default) and xhigh; Hydrogen the same as Neon.Source: aistudio.google.com · checked 2026-11-21
  • Google’s "What’s new in Gemini 4 Argon" page says Argon has a unified 1M-token context window, long decode up to 1M tokens and tunable thinking levels low, medium and high, with medium the default, and describes it as the efficient Gemini 4 frontier model with faster responses and lower token spend per turn than Carbon.Source: aistudio.google.com · checked 2026-11-21
  • OpenRouter’s public model list (November 21, 2026) has google/gemini-4-carbon at $2 / $10, google/gemini-4-neon at $3 / $15, google/gemini-4-hydrogen at $1.50 / $7.50 and google/gemini-4-argon at $2 / $10 per million input / output tokens, each with a 1,048,576-token context window. All four can be run in the chat on this site.Source: openrouter.ai · checked 2026-11-21 · second-hand

Side by side

The four Gemini 4 models Google lists: Neon, Hydrogen, Carbon and Argon

Google’s model list shows four Gemini 4 models, and all four share a 1M-token context window and the same input types. They differ in release date, price, how much they think per turn and where Google sells them. Hydrogen is the cheapest and the only one on the free API tier, Neon sits between Hydrogen and Carbon, and Carbon and Argon launched earlier and share a $2 / $10 price that runs through November 30. Each has its own page here: Gemini 4 Neon, Gemini 4 Hydrogen, Gemini 4 Carbon and Gemini 4 Argon. A name that is not in the table, Gemini 4 Krypton, is a rumour. For the app and older models see Google Gemini.

Our table, from Google’s Gemini API model list, release notes, pricing page, "Choosing a model" page and "What’s new in Gemini 4 Argon" page, and OpenRouter’s model list, read on November 21, 2026; prices are per million input / output tokens and credits are our arithmetic · checked 2026-11-21
ModelGoogle GA dateGoogle price (input / output)Thinking levels (default)Free API tierRun it here
Gemini 4 Neon2026-11-18$3 / $15low, medium, high (default), xhighNoYes: google/gemini-4-neon, about 4 credits per typical reply (our estimate)
Gemini 4 Hydrogen2026-11-12$1.50 / $7.50low, medium, high (default), xhighYesYes: google/gemini-4-hydrogen, about 2 credits per typical reply (our estimate)
Gemini 4 Carbon2026-11-10$2 / $10 through Nov 30; $4 / $20 from Dec 1low, medium (default), highNoYes: google/gemini-4-carbon, about 3 credits per typical reply at OpenRouter’s $2 / $10 (our estimate)
Gemini 4 Argon2026-10-13$2 / $10 through Nov 30; $4 / $20 from Dec 1low, medium (default), highNoYes: google/gemini-4-argon, about 3 credits per typical reply at OpenRouter’s $2 / $10 (our estimate)

Source: aistudio.google.comSource: aistudio.google.comSource: aistudio.google.comSource: openrouter.aiSource: aistudio.google.com

Our table

How the Gemini 4 models are related: which one is built from which

  • Google’s Gemini 4 Argon model card says Argon is a new frontier model, not a modification or fine-tune of an earlier one, and describes it as a large-scale sparse mixture-of-experts transformer with native multimodal support for text, vision and audio inputs. It gives a March 2026 knowledge cutoff for Gemini 4, with some domains possibly less up to date, and lists the Gemini app, Gemini API, Google AI Studio, Gemini for business, Google Antigravity and Google AI Mode as channels.Source: deepmind.google · checked 2026-11-21
  • Google’s Gemini 4 Carbon model card says Carbon is based on Gemini 4 Argon and aimed at long-horizon agentic work in knowledge work, engineering, cybersecurity and scientific reasoning. It lists the Gemini app, Gemini API, Google AI Studio, Gemini for business, Google Antigravity, Google AI Mode and the Fairwind Program for verified cybersecurity defenders as channels, and says access varies by channel and subscription tier.Source: deepmind.google · checked 2026-11-21
  • Google’s Gemini 4 Hydrogen model card says Hydrogen is adapted from Gemini 4 Carbon and built for rapid execution across agentic workflows, while Carbon and Argon are built for deep, long-horizon reasoning. It lists the Gemini app, Gemini API, Google Sheets, Google AI Studio, Gemini for business, Google Antigravity and Google Search as channels.Source: deepmind.google · checked 2026-11-21
  • Google’s Gemini 4 Neon model card (read November 19, 2026) says Neon is adapted from Gemini 4 Carbon and "balances depth and efficiency" between Carbon, which takes the hardest long-horizon projects, and Hydrogen, which handles fast everyday execution.Source: deepmind.google · checked 2026-11-19
  • Google’s post "Introducing Gemini 4 Neon" (November 18, 2026) says Neon is rolling out in the Gemini API in Google AI Studio, in Gemini for business, in Google Antigravity for paid Google AI Pro and Ultra subscribers and in the Gemini app for Pro and Ultra subscribers, and describes it directing parallel Hydrogen subagents in Antigravity.Source: blog.google · checked 2026-11-21

Family tree

What each of the Gemini 4 models is built from, and who it is for, by Google’s model cards

Google’s model cards give the Gemini 4 line a clear order: Argon first, a new model, then Carbon built on it, then Hydrogen and Neon adapted from Carbon. That is why Carbon and Argon share a price and Hydrogen is the small, fast one. It is also why the order of release (Argon, Carbon, Hydrogen, Neon) is not the order of size. For the dates and prices of each see Gemini 4 Argon and Gemini 4 Carbon.

Our table, from Google DeepMind’s Argon, Carbon and Hydrogen model cards and Google’s Neon post and model card; the Neon card was read on November 19, 2026 and the others on November 21, 2026 · checked 2026-11-21
ModelBuilt from (Google’s model card)What Google says it is for
Gemini 4 ArgonA new frontier model, not a modification or fine-tune of an earlier one; sparse mixture-of-expertsDeep work in coding, finance, legal, cybersecurity and science; the efficient frontier model, faster than Carbon
Gemini 4 CarbonBased on Gemini 4 ArgonLong-horizon agentic work, deep verification and security audits
Gemini 4 HydrogenAdapted from Gemini 4 CarbonRapid execution: everyday coding, agent workflows, parallel subagents
Gemini 4 NeonAdapted from Gemini 4 Carbon (model card read November 19)Complex, interactive software engineering, knowledge work and science, between Carbon and Hydrogen in depth and speed

Source: deepmind.googleSource: aistudio.google.comSource: deepmind.googleSource: aistudio.google.comSource: deepmind.googleSource: deepmind.googleSource: blog.google

Our table

Gemini 4 models benchmarks: Carbon against Argon, GPT-6 Astra and Claude Opus 5.5

  • Google’s Gemini 4 Carbon model card lists results as of October 2026 for Carbon against Gemini 4 Argon, GPT-6 Astra and Claude Opus 5.5; for example Terminal-Bench 4.0 (AA) 65.2% for Carbon against 57.1% for Argon, and GraphWalks from 256k to 1M tokens 96.2% against 84.2%. These are Google’s own figures; we did not run any of the tests.Source: deepmind.google · checked 2026-11-21

Maker-reported

Gemini 4 models: eight rows from Google’s Carbon model card, results as of October 2026

Google’s own card puts Carbon ahead of Argon on every row below, and ahead of both rivals on six of the eight. It trails Claude Opus 5.5 on OSWorld-2.1 and GPT-6 Astra on FrontierSWE 2, which shows these are not a clean sweep. The tests, versions and dates differ from card to card, so a Hydrogen or Neon number on another card is not comparable with this table. Use it to see the gap between Carbon and Argon, then try both in the chat: Gemini 4 Carbon and Gemini 4 Argon run on your own prompt.

Our table of eight of the seventeen rows on Google DeepMind’s Gemini 4 Carbon model card, read on November 21, 2026; Google’s figures, not tested by us; higher is better except where the card says otherwise · checked 2026-11-21
BenchmarkGemini 4 CarbonGemini 4 ArgonGPT-6 AstraClaude Opus 5.5
AA Index v5.059505758
Vals Index v2.270.4%69.2%65.0%70.2%
SWE-sweep9.6%4.3%9.5%3.6%
FrontierSWE 261.2%55.0%65.5%62.3%
Terminal-Bench 4.0 (AA)65.2%57.1%59.1%59.6%
OSWorld-2.1 (offline subset)81.5%74.7%80.9%87.2%
GraphWalks 256k to 1M, BFS (F1)96.2%84.2%71.8%66.8%
RiemannBench80.0%76.0%72.0%69.6%

Source: deepmind.google

Nearest available

Alternatives to Gemini 4 models you can use now

$1.5 / $7.5 (per million tokens)

  • Google announced Gemini 4 Hydrogen on November 12, 2026 (the post was last updated November 16, 2026), and its model card is dated the same day.Source: blog.google · checked 2026-11-17
  • The Gemini API changelog lists gemini-4-hydrogen as generally available (GA) on November 12, 2026, with a new xhigh thinking level and as the default model for the Antigravity agent in Gemini Managed Agents; it points to migration guidance from Gemini 3.8 Flash.Source: aistudio.google.com · checked 2026-11-17
  • Google describes Gemini 4 Hydrogen as a fast, efficient model for everyday coding and agentic work, built for "rapid execution across diverse agentic workflows". Gemini 4 Carbon and Gemini 4 Argon remain its frontier models for long-horizon reasoning.Source: blog.google · checked 2026-11-17

$3 / $15 (per million tokens)

  • Google announced Gemini 4 Neon on November 18, 2026 in a post by Koray Kavukcuoglu (SVP, Google DeepMind and Chief AI Architect). It describes Neon as "frontier-level intelligence with fast end-to-end task completion" across software engineering, knowledge work and science. The model card is dated the same day.Source: blog.google · checked 2026-11-19
  • The Gemini API release notes list gemini-4-neon as generally available (GA) on November 18, 2026, on the paid tier at $3.00 per million input tokens and $15.00 per million output tokens, with four thinking levels (low, medium, default high, xhigh), a unified 1M-token context window and long decode up to 1M tokens.Source: aistudio.google.com · checked 2026-11-19
  • Google’s deprecations page lists gemini-4-neon with release date November 18, 2026 and "No shutdown date announced". Google Cloud’s model page gives its launch stage as GA.Source: aistudio.google.com · checked 2026-11-19

$2 / $10 (per million tokens)

  • Announced on October 28, 2026 for a subset of cyber defenders, with developers and consumers to follow.Source: blog.google · checked 2026-11-11
  • The Gemini API changelog lists gemini-4-carbon as generally available (GA) on November 10, 2026.Source: aistudio.google.com · checked 2026-11-11
  • Google’s model card lists multi-step coding, scientific research, knowledge work and cyber defense.Source: deepmind.google · checked 2026-11-11

$2 / $10 (per million tokens)

  • Google announced Gemini 4 Argon on September 30, 2026, saying it was rolling out to trusted cyber defenders through its Fairwind Program, with release to developers, enterprises and consumers to follow "as soon as possible", starting with paid API customers and Google AI Ultra subscribers. The post gives no date and does not say how long the introductory price lasts.Source: blog.google · checked 2026-11-16
  • The Gemini API changelog lists gemini-4-argon as generally available (GA) on October 13, 2026.Source: aistudio.google.com · checked 2026-11-16
  • Google’s model card (published October 13, 2026) lists these channels: the Gemini app, Gemini API, Google AI Studio, Gemini for business, Google Antigravity and Google AI Mode. Google’s pricing page says Gemini 4 Argon is not available on the Free tier.Source: deepmind.google · checked 2026-11-16

FAQ

Gemini 4 models FAQ

What are the Gemini 4 models?

Google’s Gemini API model list has an "All Gemini 4 models" table with four rows: Gemini 4 Carbon, Neon, Hydrogen and Argon, with the model codes gemini-4-carbon, gemini-4-neon, gemini-4-hydrogen and gemini-4-argon. Argon was released on October 13, 2026, Carbon on November 10, Hydrogen on November 12 and Neon on November 18. All four have a 1M-token context window and take text, image, audio, video and PDF input. Each has its own page here: Gemini 4 Argon, Gemini 4 Carbon, Gemini 4 Hydrogen and Gemini 4 Neon.

Which Gemini 4 model should I use?

Google’s "Choosing a model" page recommends Hydrogen for everyday coding and agent work, Neon for complex software engineering and analysis where you are waiting on the answer, and Carbon for long background runs, deep research and security audits. Argon is not in that table; Google describes it as the efficient frontier model, with faster responses and lower token spend per turn than Carbon. The cheapest way to find out is to run your own prompt on two of them in the chat.

How much do the Gemini 4 models cost?

Google’s Gemini API pricing page lists, per million input / output tokens, Hydrogen at $1.50 / $7.50, Neon at $3 / $15, and Carbon and Argon at $2 / $10 through November 30, 2026, rising to $4 / $20 from December 1. Only Hydrogen is on the free API tier. On this site a typical reply costs about 2 credits with Hydrogen, 3 with Carbon or Argon and 4 with Neon (a credit is about $0.01 in the $5 pack), billed by real token cost, at least 1 credit; these four need credits you bought, because the sign-up gift credits cover only models up to $1 / $5. Prices and limits can change, so check Google’s pages.

How are the Gemini 4 models related?

Google’s model cards say Argon is a new model, not a modification of an earlier one, Carbon is based on Argon, and Hydrogen and Neon are adapted from Carbon (we read the Neon card on November 19). So Carbon and Argon share a price, Hydrogen is the small, fast one, and Neon sits between Hydrogen and Carbon in depth and speed. The release order is therefore not the size order.

Which Gemini 4 model scores best?

On Google’s own Carbon model card (results as of October 2026), Carbon is ahead of Argon on all eight rows in our table, for example Terminal-Bench 4.0 (AA) 65.2% against 57.1% and GraphWalks from 256k to 1M tokens 96.2% against 84.2%, but it trails Claude Opus 5.5 on OSWorld-2.1 (81.5% against 87.2%) and GPT-6 Astra on FrontierSWE 2 (61.2% against 65.5%). Google’s Hydrogen card shows 58.8% on HLE-Diamond against 53.3% for Claude Sonnet 5.5, and its Neon post claims 61 on the Artificial Analysis Intelligence Index. These are Google’s figures from different cards, not tested by us.

Is there a Gemini 4 Krypton, Flash or Pro?

Not on Google’s list: the "All Gemini 4 models" table has only Carbon, Neon, Hydrogen and Argon, with no row called Krypton, Flash or Pro. Krypton is a leaked name that Google has not confirmed; see Gemini 4 Krypton. The earlier Flash line is covered on Gemini 3.8 Flash, and the app, plans and older models on Google Gemini.

Can I try the Gemini 4 models without a Google account?

Yes: all four are on OpenRouter’s model list and run in the chat on this site, with a model picker and one credit balance, no Google account needed; you sign in to this site, and these four use credits you bought. The Gemini app and Google’s own plans are not offered here.

Is this the official Google website?

No. This page is unofficial and not affiliated with or endorsed by Google. Facts link to the pages we read.