Photograph: Google (press image)Google makes Gemini Tasks free in six countries
The rebranded Spark agent runs on Gemini 4 Hydrogen, with calls, payments and messaging available only in the US.
Jules Park, Nov 17, 2026
The new model costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window, and is available now to paid Gemini subscribers.

Google said in a blog post on Wednesday that Gemini 4 Neon is rolling out now for complex software engineering, knowledge work and science workflows. The model is available in the Gemini app for paying Google AI Pro and Ultra subscribers, in Google Antigravity for those subscribers, through the Gemini API in Google AI Studio and Android Studio, and through Gemini for business. It has a 1 million-token context window.
The model costs $3 per million input tokens, $0.15 per million cached input tokens, and $15 per million output tokens. 9to5Google reported that this is twice the price of Gemini 4 Hydrogen, and after promotional pricing for Argon and Carbon ends on November 30, Neon will be 25 percent cheaper than those models' $4 input and $20 output rates. 9to5Google also reports four thinking levels: Low, Medium, High and Extra High.
Google cites independent evaluations that put Neon third overall on the Artificial Analysis Intelligence Index with a score of 61 at its xhigh thinking level, second on the Vals Index at 70.9 percent, and second on the new Agent Arena Index with a score of 1,336 across more than 4,200 real-world agent sessions. Google also says Neon ranks first on Legal Research Bench, Tax Agent Bench, Zapier AutomationBench at 56.3 percent, Andon Labs Vending-Bench 2 at $16,843, and the Artificial Analysis Coding Agent Index at 58.6.
At its default high thinking level, Google says Neon costs 32 percent less per task than Carbon, while at xhigh it still costs at least 36 percent less per task than competing models at their highest setting. In the Gemini app, Google says Neon writes more clearly, reads uploads such as spreadsheets, PDFs and photos more precisely, and pulls in current web details for comparisons like vendor pricing.
9to5Google also reported that the app's model picker now shows Auto, Everyday help (4 Hydrogen), Deeper analysis (4 Neon), and Code & complex work (4 Carbon), with Gemini 3.1 Pro going away, Neon replacing Argon, and 3.5 Flash-Lite becoming Auto.
Google says the Neon for cyber variant produces more secure code than Carbon across all AutoBaxBench challenge levels, reaching 76 percent on the hard tier, and hits 51.26 percent on WebExploitBench v1.2 at half the cost of competing large models. Eligible Fairwind Program defenders will get access in the coming days.
Sources: Google, 9to5Google
Photograph: mliu92 from San Mateo/Wikimedia Commons (CC BY-SA 2.0)
Photograph: Asoundd/Wikimedia Commons (CC BY-SA 4.0)
Photograph: Rockstar Games (press image)
Photograph: Nvidia (press image)
Photograph: Nvidia (press image)