DEV Community

Trevor H
Trevor H

Posted on Originally published at trevorheshka.com

New Google AI Models: Oct-Nov 2026 Guide

These are the new Google AI models released from mid-October to November 21, 2026: four Gemini 4 models, plus world, video, decision and robotics models. Here is what each is best at, where to get it, what it costs, and what it means for developers, small businesses and IT teams.

Prices are in US dollars per million tokens, from Google's own pages.

The Gemini 4 lineup at a glance

Model Released Best at API price (in / out) Free tier
Gemini 4 Argon Oct 13 Deep multi-step work in coding, finance, legal, security and science $2 / $10 until Nov 30, then $4 / $20 No
Gemini 4 Carbon Announced Oct 28, GA Nov 10 Cybersecurity defense and long-running coding and research $2 / $10 until Nov 30, then $4 / $20 No
Gemini 4 Hydrogen Nov 12 Fast everyday coding and agent tasks $1.50 / $7.50 Yes
Gemini 4 Neon Nov 18 Complex coding, analysis and business workflows, done fast $3 / $15 No

All four have a 1 million-token context window.

Gemini 4 Argon: the first Gemini 4 model

Best at: deep, multi-step work in coding, finance, legal, cybersecurity and science.

Gemini 4 Argon launched globally on October 13. Google says it ranked #1 on the Vals Index. It is on the paid Gemini API, Gemini for business, Antigravity (AI Ultra), and the Gemini app and AI Mode in Search (AI Pro and Ultra). Cached input is 95% off.

For IT teams: Google reports Argon cut attack success on the Gray Swan indirect prompt injection benchmark to 0.7%. Google Cloud also offers gemini-4-argon-cyber-permissive for defensive security work (Agent Platform release notes).

Gemini 4 Carbon: the security-heavy frontier model

Best at: cybersecurity defense (finding and patching vulnerabilities) and long-running coding, debugging and research jobs.

Google announced Carbon on October 28 as its strongest model, first releasing it only to cyber defenders and U.S. government testers. Google reported 88.2% on SEC-bench Pro (finding vulnerabilities) and 71.0% on CWE-bench (patching them).

On November 10 it opened to paying customers: the paid Gemini API, Gemini for business and AI Ultra subscribers. A defensive variant, gemini-4-carbon-cyber-permissive, sits behind the Fairwind allowlist on Google Cloud.

Watch the price. Like Argon, the $2 / $10 rate ends November 30, then becomes $4 / $20. Carbon also uses more tokens than Argon, so test real jobs before you budget.

For IT teams: keep a human approving every fix.

Gemini 4 Hydrogen: the one most people will use

Best at: fast, cheap everyday coding and agent tasks, plus spatial reasoning for robots.

Gemini 4 Hydrogen launched November 12. It is the only Gemini 4 model on the API free tier and replaces Gemini 3.8 Flash. It is also in the Gemini app, Search, Sheets and Gemini for business. The Gemini API changelog lists four thinking levels: low, medium, high (default) and a new xhigh. On November 17 Google showed it driving robot arms via an Antigravity plugin.

For small businesses: start here for email drafts, spreadsheets or a support chatbot.

Gemini 4 Neon: the middle option

Best at: complex coding, analysis and business workflows that run often and need to finish quickly.

Gemini 4 Neon arrived November 18, between Hydrogen and Carbon. Google says it costs 32% less per task than Carbon. It is on the paid API ($3 / $15, cached input $0.15), Gemini for business, and the Gemini app and Antigravity for AI Pro and Ultra subscribers. 9to5Google reports it replaces Gemini 3.1 Pro in the app.

Per token it costs more than Argon's and Carbon's intro rates; the savings come from using fewer tokens.

Genie 4: interactive worlds with sound

Best at: real-time interactive world simulation with sound, for prototypes and for training robots and AI agents.

DeepMind announced Genie 4 on October 27 as a research preview. It builds a world from text or an image that you explore live. It reached AI Ultra subscribers in Project Genie on November 18 and the Gemini API on November 19: 720p at 24 fps, steered with WASD or arrow keys and text, at about $0.10 per second on the paid tier.

Details drift in longer sessions, so don't use it for production video.

Gemini Omni 1.2 Flash: cheaper business video

Best at: short videos with sound at scale, such as product clips and ads, with characters and products kept consistent.

Gemini Omni 1.2 Flash went GA on November 5. It makes clips up to 15 seconds, edits them through follow-up prompts, and takes up to six reference images. The price dropped 30% to $0.07 per second at 720p. It is in the Gemini API, Google Cloud, Flow, Vids and Ads.

Migration note: Veo 3.1 previews shut down October 22 (Omni 1.2 Flash replaces them), and gemini-omni-1.1-flash routes to 1.2 on December 3 (deprecations page).

Gemini Decision 1: yes/no answers without the chatter

Best at: fast yes/no, multiple-choice and rubric decisions, such as routing, tool selection and policy checks.

Gemini Decision 1 launched November 11 with a new Decisions API. Rather than writing text, it scores the options and returns a confidence for each. It takes up to 1,000 questions per input, including images, audio, video and PDFs. Google reports 150-250 ms end to end, at $0.10 per million input tokens and nothing for output.

For developers and IT: ticket routing and phishing triage no longer need a custom classifier.

Gemini Robotics 2.5

Best at: controlling robot arms and hands for dexterous physical tasks.

Gemini Robotics 2.5, announced November 10. In a Mercedes-Benz production-line pilot, an Apptronik Apollo 2 humanoid sorted parts with a 96% success rate. It is for robotics partners only.

Nothing new on these

No new Gemma, Imagen, Lyria, Veo, MedGemma or Gemini Nano models in this window. Nano Banana 2.1 (released October 6) started rolling into Google Pics, Docs and Slides on November 16, and the EmbeddingGemma 2 technical report followed November 17.

What to do this week

  • Check old model strings. On December 1, gemini-2.5-pro and gemini-2.5-flash reroute to newer models, with breaking API changes (changelog).
  • Budget for Argon and Carbon doubling to $4 / $20 on December 1.
  • Start new projects on Hydrogen and move up only when needed.
  • Try the Decisions API for yes/no steps.

FAQ

What new Google AI models came out in October and November 2026?
Gemini 4 Argon (October 13), Gemini Omni 1.2 Flash (November 5), Gemini 4 Carbon (GA November 10), Gemini Robotics 2.5 (November 10), Gemini Decision 1 (November 11), Gemini 4 Hydrogen (November 12), Gemini 4 Neon (November 18) and Genie 4 (Project Genie and the Gemini API, November 18-19).

Which Gemini 4 model should a small business use?
Gemini 4 Hydrogen. It is on the free API tier and costs $1.50 / $7.50 on the paid tier.

How much do Gemini 4 Argon and Carbon cost?
Both are $2 in / $10 out per million tokens until November 30, 2026, then $4 / $20. Neither is on the free tier.

Is Gemini 2.5 going away?
On December 1, 2026, API calls to gemini-2.5-pro reroute to Gemini 3.8 Flash and gemini-2.5-flash to Gemini 3.5 Flash-Lite, billed at those rates. Update your code before then.

Top comments (0)