Google Launches Gemini 4 Neon for AI Pro Subscribers

Google is rolling out Gemini 4 Neon today in the Gemini app for AI Pro and AI Ultra subscribers. The model demonstrates comparable, if not superior, performance to Carbon for coding, reasoning, and agents, with speed similar to Flash and at a lower cost.
Neon Will Be Cheaper After the Promotion
Neon's pricing is set at $3/1M tokens received and $15/1M tokens generated. This is double that of Hydrogen. Until November 30, Argon and Carbon remain priced lower: $2/1M for inputs and then $10/1M for outputs.
Once this promotion ends, the rates for Argon and Carbon will increase to $4/1M for inputs and $20/1M for outputs. Neon will then be 25% cheaper than these models.
Subscribers See Their Selector Change
In Gemini, Argon is replaced by Neon, while version 3.1 is removed. The menu now associates “In-depth Analysis” with 4 Neon, “Daily Help” with 4 Hydrogen, and “Code and Complex Tasks” with 4 Carbon.
For paid accounts, 3.5 Flash-Lite appears under the title “Auto,” presented by Google as a choice that adjusts to needs. Gemini tasks can also operate with Neon. The model is also accessible from Google Antigravity for holders of a paid AI Pro or Ultra subscription.
Google Targets Neon for Complex Professional Work
Following the releases of Carbon and Hydrogen last week, Google positions Neon between these models in Gemini 4. Hydrogen, the successor to Flash, remains aimed at fast daily activities; Carbon targets the most demanding long-term projects, pushing the limits of the model's capabilities. Argon was previously offered for analysis and development, with Google recommending it for most software or knowledge work.
Neon is designed for complex development, analytical reasoning, and production business processes. Its context window reaches 1 million tokens. Like Hydrogen, it features four reflection settings: Low, Medium, High, and Very High.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
For writing, Google announces sharper and better-organized responses, with a direct and natural tone. The company also claims a more accurate interpretation of imported tables, PDFs, and photos, transforming dense content into summaries or reports, as well as extracting relevant information from Gmail or Docs. According to Google, Neon regularly leverages the Web to update its responses, particularly when comparing prices between suppliers before a significant purchase or preparing a summary that incorporates recent market data.
Google also asserts that complex multi-step tasks are executed faster and more efficiently than with larger models. The company promises more delegations before reaching usage caps. It cites reviewing invoices to explain a fee increase or reconciling receipts and documents to negotiate a refund.
Neon Scores 61 in Artificial Analysis
In the Artificial Analysis Intelligence Index, Neon ranks 3rd with a score of 61 at the xhigh setting, ahead of comparable models but behind larger and slower models. Artificial Analysis also ranks it 1st in its Coding Agent Index, with a score of 58.6. Neon achieves 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0, the latter result corresponding to the high setting offered by default.
At Arena, Neon ranks 2nd globally in the Agent Arena Index: its score reaches 1336 across more than 4,200 real agent sessions. It also ranks 2nd for its ability to account for user corrections during a task. On Agents’ Last Exam v1.1, it scores 75.8%, a result on par with leading models.
Neon ranks 1st in legal research and tax agent evaluations. On Surge AI, it comes 1st in most financial, accounting, table modeling, and complex constraint compliance tests. For Chartography, dedicated to dense professional charts, it scores 73.5%. Neon is also presented as the best in its category on Terminal-Bench Science with 70.5%, RiemannBench with 84.0%, and HLE-Diamond with 71.3%.
In AutomationBench, Neon ranks 1st overall with 56.3%. It also finishes 1st on Vending-Bench 2 from Andon Labs, a long-duration business management simulation, with a result of $16,843.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.