Gemini 4 Neon Slots Between Carbon and Hydrogen at $3 per Million Input Tokens, With Google Claiming a 32% Lower Cost per Task Than Carbon

Google released Gemini 4 Neon on November 18, 2026, a mid-tier model in the Gemini 4 family priced at $3/$15 per million tokens with a 1 million-token context window. Google says it ranks third on the Artificial Analysis Intelligence Index, second on the Vals Index and first on the Artificial Analysis Coding Agent Index, and that a cyber variant is coming to vetted defenders.

A glowing orange-red neon discharge tube stands on a dark lab bench between a rough black carbon block and a small clear hydrogen flask.
A glowing orange-red neon discharge tube stands on a dark lab bench between a rough black carbon block and a small clear hydrogen flask.

Google has filled the middle of its Gemini 4 lineup. On November 18, 2026, the company introduced Gemini 4 Neon, a model it describes as bringing “frontier-level intelligence with fast end-to-end task completion” to software engineering, knowledge work and science. Neon arrives eight days after Gemini 4 Carbon opened to paying developers and six days after the faster Gemini 4 Hydrogen.

Google positions the three models plainly: Carbon handles the hardest, long-horizon projects, Hydrogen handles fast everyday execution, and Neon is meant to balance depth and efficiency for “complex, high-frequency workflows.”

Pricing that lands between its siblings

Neon is priced at $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million, a 95% discount. It has a 1 million-token context window.

That places it above Hydrogen, which Google prices flat at $1.50 input and $7.50 output across its full context window. The comparison with Carbon depends on timing. Carbon is currently on an introductory price of $2 and $10 per million tokens, and Google says that rises to $4 and $20 for the public-access Carbon model after November 30, 2026. Per token, Neon is therefore more expensive than Carbon this month and cheaper from December.

Google’s argument rests on cost per task rather than per token. At its default “high” thinking level, Google says Neon pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon, and sits on the Pareto frontier of the Vals Index and Agent Arena at roughly 50% to 80% lower cost per task than top competing frontier models. Neon has four thinking levels; at the top “xhigh” setting, Google says it still costs at least 36% less per task than competing models at their highest settings.

Where Google says it ranks

At xhigh, Neon scores 61 on the Artificial Analysis Intelligence Index, which Google says puts it third overall, behind larger and slower models. Google lists a set of third-party rankings alongside its own figures:

  • Vals Index: second, at 70.9%, and first on the Legal Research Bench and Tax Agent Bench
  • Artificial Analysis Coding Agent Index: first overall, at 58.6
  • Agent Arena Index: second globally, with a score of 1336 across more than 4,200 real-world agent sessions, and second on steerability
  • Zapier’s AutomationBench: first, at 56.3%
  • Andon Labs’ Vending-Bench 2: first, with $16,843 in its long-horizon business simulation

On coding, Google gives 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at the default high setting. For science and mathematics, Google cites 70.5% on Terminal-Bench Science, 84.0% on RiemannBench and 71.3% on HLE-Diamond. It also reports 75.8% on Agents’ Last Exam v1.1, which it describes as on par with leading models.

All of these figures come from Google’s launch post. Several are third-party leaderboards, but the selection of which ones to highlight is Google’s.

Built to direct other agents

Google says Neon is designed for long-running, tool-heavy agent workflows. In Antigravity, it can run on its own or act as an orchestrator that directs parallel Hydrogen subagents. The company says Neon cuts redundant file reads, reports progress as it works and handles context compaction reliably in long sessions. It also claims Neon recorded Google’s lowest fabrication rate yet in internal Antigravity testing.

In the Gemini app, Google says Tasks in Gemini can now run on Neon.

A cyber variant for vetted defenders

As with Carbon, Google is preparing a security version. Gemini 4 Neon for cyber reaches 76% on the hard tier of AutoBaxBench for secure code generation, 51.26% on WebExploitBench v1.2, 90.5% on SEC-bench Pro for vulnerability finding and 70.7% on CWE-bench v2 for patching, according to Google. It is rolling out to eligible members of Google’s Fairwind Program “in the coming days,” with access requested through a Cloud account team.

On safety, Google says Neon keeps Carbon-level protections against cyber and biosecurity misuse while refusing just 1.98% of tasks on Terminal-Bench 4.0 and none on Terminal-Bench-Science. It reports a 0.0% cheating rate on Terminal-Bench 4.0, shortcuts on 2.45% of tasks in its coding cheating evaluation, a 0.09% misaligned action rate on its agentic safety suite and a 99.2% pass rate on its agent honesty evaluation. With real-time safeguards at its highest thinking level, Google says Neon resisted every attack in Gray Swan’s indirect prompt injection benchmark.

Rolling out in AI Studio, Antigravity and the Gemini app

Neon is rolling out in the Gemini API paid tier in Google AI Studio and Android Studio, in Gemini for business through the API and model menu, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app model menu for Pro and Ultra subscribers. Google has published a separate model card with further detail.

Topics
  • #Google
  • #Gemini 4 Neon
  • #Gemini
  • #Model Launch
  • #Google Antigravity
Raj M

Author

Raj M

Contributor

AI Systems Architect is a seasoned technology leader with over 15 years of experience in the IT industry working with Fortune 500 companies. With a solid foundation in multi-agent systems, open-source LLM infrastructure, and enterprise deployment, he excels at building scalable production-grade AI platforms.