announcementsai-codecomparison

Gemini 4 Neon costs $3/$15, double Hydrogen. From December 1 it undercuts Argon and Carbon by 25%.

Google's fourth Gemini 4 model is in the API, Antigravity and the Gemini app for AI Pro and Ultra, with no free API tier. Where it fits next to Hydrogen, Argon and Carbon, and how it compares with GPT-6.2 Sol and Claude.

November 20, 2026

Gemini 4 Neon costs $3/$15, double Hydrogen. From December 1 it undercuts Argon and Carbon by 25%.

Google released Gemini 4 Neon on November 18, 2026, its fourth Gemini 4 model in seven weeks after Argon (September 30), Carbon (October 28) and Hydrogen (November 12). Neon costs $3 per million input tokens and $15 per million output tokens on the paid Gemini API tier, with no free tier. It is also in the Gemini app and Google Antigravity for Google AI Pro and Ultra subscribers, and in Gemini for business.

Where Neon fits, and what happens to Argon

Google's post: "While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency." The model card says Neon is adapted from Carbon, with a March 2026 knowledge cutoff. Paid-tier prices per million tokens, from Google's pricing page:

  • Gemini 4 Hydrogen: $1.50/$7.50, the only Gemini 4 model on the free tier.
  • Gemini 4 Neon: $3/$15, with no introductory price.
  • Argon and Carbon: $2/$10 through November 30, then $4/$20.

So Neon costs twice Hydrogen, 50% more than Argon and Carbon until the end of the month, and 25% less than them from December 1.

Google's announcement does not say Neon replaces Argon. In the Gemini app, though, Argon is gone: Google's model help page lists Hydrogen and Neon for AI Pro, adds Carbon for Ultra, and no longer mentions Argon. 9to5Google writes: "Neon also replaces Argon." In the API, the models page still lists gemini-4-argon, and the changelog has no deprecation notice for it.

API details: no free tier, four thinking levels, 1M context

  • Model ID: gemini-4-neon, generally available from day one.
  • Free tier: none; the pricing page marks Neon "not available" there. Google says AI Pro and Ultra subscribers can try it in AI Studio.
  • Discounts: cached input costs $0.15 (implicit caching only). Batch and Flex cost $1.50/$7.50, Priority $5.40/$27.
  • Context: one 1,048,576-token window shared by input and output. Responses stop at 65,536 tokens unless you use long decode.
  • Thinking: low, medium, high (the default) and xhigh, as on Hydrogen.
  • Tools: text, image, audio, video and PDF input, text output. The model page lists function calling, computer use, code execution and Search grounding, but no Live API.

What changes in the Gemini app and Antigravity

Gemini app. Per Google's help page, users without a plan get Auto (Hydrogen or a lighter model, depending on the prompt), AI Plus ($4.99 a month) adds Hydrogen, AI Pro ($19.99) adds Neon in chats and Tasks, and AI Ultra ($99.99) adds Carbon. 9to5Google reports the picker labels: Auto, "Everyday help" for 4 Hydrogen, "Deeper analysis" for 4 Neon and "Code & complex work" for 4 Carbon, and says Gemini 3.1 Pro is "finally going away." Google has not published a Neon-specific usage limit.

Antigravity. Per the Antigravity post, Neon is rolling out to paid Pro and Ultra subscribers in the desktop app, CLI and IDE extensions. Orchestrator can now pair Neon as planner with Hydrogen as executor in one session: "If Hydrogen hits friction, it hands back to Neon to inspect state, repair the plan, and resume execution." At Hydrogen's launch, Pro got an Orchestrator built on Argon (see our Hydrogen post); Google has not said whether it stays.

Google's benchmarks, and an outside check

These are Google-reported. The model card compares Neon with Claude Opus 5.5 and GPT-6.2 Sol, in that order:

  • Artificial Analysis Intelligence Index: 61, 59, 57.
  • Vals Index v2.2: 70.9%, 70.1%, 66.1%.
  • Terminal-Bench 4.1 (cheating-adjusted): 65.2%, 65.8%, 61.5%.
  • FrontierCode 2.0: 46.2%, 47.3%, 43.8%.
  • Zapier AutomationBench: 56.3%, 42.5%, 41.3%.
  • Chartography: 73.5%, 66.3%, 75.1%.

Opus 5.5 leads on the two coding rows and Sol on charts. Neon's widest margin, 39.0% on SWE-sweep against 10.1% and 8.3%, comes from Google's own Antigravity harness; on the mini-SWE harness it is 12.8%, 3.6% and 11.6%. The model card also says Neon "reduces unjustified refusals, particularly on multimodal prompts."

Google says Neon at high costs 32% less per task than Carbon. Artificial Analysis, quoted by TestingCatalog, found "a small Intelligence gain over Gemini 4 Carbon at a ~17% lower Cost per Task." Its model page scores Neon 61 at xhigh and 60 at high, against 60 for Carbon, 54 for Hydrogen and 50 for Argon. Xhigh used 283M output tokens on the index, more than twice the 123M at high. Our leaderboard tracks outside results.

Which Gemini 4 model to use, and how Neon's price compares

List prices per million tokens: GPT-6.2 Sol $2/$10 on OpenAI's pricing page ($4/$15 above 272K tokens), Claude Sonnet 5.5 $2/$10 and Opus 5.5 $4/$20. For a month with 200 million input and 20 million output tokens, before caching:

  • Hydrogen: $450.
  • Sol, Sonnet 5.5, or Argon and Carbon through November 30: $600.
  • Neon: $900.
  • Opus 5.5, or Argon and Carbon from December 1: $1,200.

Per task, the order can change. Arena, quoted by TestingCatalog, put Neon at "$2.82 per task" on its Agent Arena Index: "That is 53% less than Claude Opus 5.5, which Neon outranks." The pricing index covers other vendors.

  • On Argon via the API: it stays cheaper until November 30. Test Neon now; from December 1 it costs 25% less and scores 61 to Argon's 50 on Artificial Analysis.
  • On Carbon: Neon at high matches Carbon's index score for 25% less per token from December 1. Google still points to Carbon for the hardest long-horizon work (see our Carbon post).
  • Need a free tier or the lowest Gemini price: stay on Hydrogen.
  • On Sol or Sonnet 5.5: Neon costs 50% more per token and leads Sol on five of the six rows above, so compare cost per finished task on your own work.
  • On AI Pro in the app: "Deeper analysis" is the strongest model you can pick, since Carbon needs Ultra.

Also this week, and what Google has not said

On November 19, Genie 4, Google's world model, reached the Gemini API's Interactions API; the changelog prices it at $1.50 per million input tokens and $17.50 per million video output tokens. Tool Retrieval entered preview the same day, loading tool schemas only when an agent needs them, for Gemini 3.6 and later.

Google has not said whether gemini-4-argon will be retired from the API, how much Neon use each app plan includes, or whether Neon will get a free API tier. The full Safety and Security Report "will be available in the coming days," per the model card, and its benchmark table leaves out Sonnet 5.5, Argon and Carbon.

Some links in this article are affiliate links. Learn more.