Google on November 18, 2026 introduced Gemini 4 Neon, a model it describes as bringing frontier-level intelligence with fast end-to-end task completion across software engineering, knowledge work and science workflows. In the Gemini 4 family, Neon sits between Gemini 4 Hydrogen, built for fast everyday execution, and Gemini 4 Carbon, aimed at the hardest long-horizon projects.
Neon is priced at $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, with a unified 1 million-token context window, according to Google's announcement and its AI Studio documentation. Those rates are 25% below Gemini 4 Carbon's standard $4 and $20, although Carbon is on promotional pricing of $2 and $10 through November 30. The model supports four thinking levels: low, medium, high (the default) and xhigh.
Independent benchmarking firm Artificial Analysis reported that Neon at its xhigh setting scores 61 on its Intelligence Index, one point ahead of Gemini 4 Carbon and third overall, behind GPT-6.2 Astra and Claude Fable 5.2 at their maximum settings. The firm measured about $6.44 per task for Neon, roughly 17% less than Carbon at standard pricing, and about 21 minutes per task versus roughly 36 for Carbon. It noted that Neon actually uses about 15% more output tokens per task, so the savings come from lower token prices and faster output, not from greater efficiency.
Google's own results span coding, enterprise work and security. The company says Neon ranks first on the Artificial Analysis Coding Agent Index at 58.6, scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default setting, and ranks second on the Vals Index at 70.9%. It also cites first-place finishes on Zapier's AutomationBench and Andon Labs' Vending-Bench 2, and state-of-the-art scores on several cybersecurity benchmarks. Apart from the Artificial Analysis measurements, these figures are supplied by Google and have not all been independently reproduced; VentureBeat noted that Neon's Vals result was provided by Google.
Availability starts immediately on the paid tier of the Gemini API in Google AI Studio and Android Studio, in Gemini for business, and for Google AI Pro and Ultra subscribers in the Gemini app and the Google Antigravity development platform. Google says a cyber-focused version of Neon will roll out to vetted defenders through its Fairwind Program over the coming days. The company also says Tasks in the Gemini app can now run on Neon.
On safety, Google says Neon keeps Carbon-level protections against severe misuse and prompt injection while declining less legitimate work, citing a 1.98% refusal rate on Terminal-Bench 4.0. It reports a 99.2% pass rate on its internal agent honesty evaluation and shortcut-taking on 2.45% of tasks in its coding cheating evaluation, the lowest of any Gemini 4 model. Those are internal evaluations described in Google's announcement and model card.
The launch extends a rapid release cycle. Artificial Analysis called Neon Google's fourth model in seven weeks, and it arrives a day after OpenAI released GPT-6.2 Astra and Sol. For developers, the practical comparison is cost per completed task: VentureBeat noted Neon's headline token prices are 50% higher than mid-range rivals such as GPT-6.2 Sol and Claude Sonnet 5.5, while Google is betting that stronger results per task justify the premium.
