Google releases Gemini 4 Neon for Chat and Tasks: 1M tokens and $3 per million input tokens
Google has added the 4 Neon model to Gemini for AI Pro and Ultra subscribers. It reads spreadsheets, PDFs and photos more accurately and pulls in emails from Gmail. In Chat and the background Tasks mode, you can hand it data exports and a contract and get a finished report back. The 1M-token context window holds several thick books, and the API charges $3 per million input tokens and $15 per million output tokens. Test it on a couple of your own documents.
AI-processed from TG: @cgevent (Метаверсище и ИИще); edited by Hamidun News
Picture a typical Monday for an analyst: three Excel exports in the inbox, a forty-page PDF contract and a photo of a whiteboard from a meeting, and one structured report due by lunch. That is the kind of work Google built Gemini 4 Neon for. In the interface, it is called "Deeper analysis."
Google has added it to the Gemini app on the web, desktop and smartphones, in both Chat (regular conversation) and Tasks (autonomous assignments that run in the background). For now, it is available to Google AI Pro and Ultra subscribers. The two figures in the headline mean the following.
"1M tokens" is the size of the context window, meaning how much text the model can hold at once: a token is a fragment of a word, and a million of them is roughly equal to several thick books. "$3 per million input tokens" is the API price for every million tokens you send to the model; for every million tokens of output, Google charges $15.
What Neon can do in regular chat
You can select the model from the dropdown list in Chat: the "4 Neon" entry. According to the Gemini App team, the main improvement concerns dense material. The model parses uploaded spreadsheets, PDFs and photos more accurately and turns them into structured reports. It also pulls context from connected services such as Gmail and Google Docs. If you ask it to combine your correspondence with a supplier and a price table, it finds the emails and documents itself, so you do not need to summarize them by hand.
Google singles out research queries separately. For fast-changing topics, such as vendor price comparisons or market summaries, Neon more often relies on fresh web search rather than on what it "remembers" from training. The writing style stays direct and clear. All of these claims come from Google itself for now. The announcement contains no independent accuracy measurements, so it is worth evaluating quality on your own documents.
Tasks: more assignments within the same limits
Tasks are multi-step jobs that the model carries out in the background while you do something else. According to Google, they run faster and more economically on Neon, so within the same Pro or Ultra subscription limits you should be able to complete more assignments. The company gives no specific speedup figures, so it is not possible to claim a difference of several times.
For a product manager, this is a reason to move routine work into Tasks: a weekly competitor summary or the review of incoming documents. You can then compare how much of your limit a single task consumes.
Technical details and pricing for developers
Neon's context window is 1M tokens. Standard output is capped at 64K tokens, and Long Decode mode raises it to 1M. There are four reasoning depth levels, so you can choose between a quick answer and a more thorough analysis. The deeper the model thinks, the longer and more expensive the response.
In the API and in Antigravity, the model costs $3 per million input tokens and $15 per million output tokens. Cached input costs $0.15 per million, which is twenty times cheaper than regular input. If you repeatedly send the same large document or a long instruction, the savings can be noticeable.
What to do with this
If you have a Pro or Ultra subscription, run three or four real work documents through Neon and compare the result with the previous model. For the API, calculate the cost of a typical request with caching factored in: on long context, caching is what determines the budget. Do not enable maximum reasoning depth by default; save it for difficult cases.
The limitations are clear: the model is available only on paid plans, the announcement shows no detailed benchmarks, and the Tasks speed gain is described only in words. It is better to decide on switching based on your own tests, not on the presentation.
- Nov 18, 2026Huang steps down as NVIDIA CEO, company loses more than $1 trillion in market cap
- Nov 18, 2026Google releases Gemini 4 Neon: 3rd place in AI ranking and tasks 32% cheaper
- Nov 18, 2026Devin Bot built an Android app on its own in 44 days: 1 134 PRs from 183 messages
- Nov 18, 2026Google finds AI agents that discover vulnerabilities and hack websites on their own
Key Questions Answered
What Neon can do in regular chat?
You can select the model from the dropdown list in Chat: the "4 Neon" entry. According to the Gemini App team, the main improvement concerns dense material. The model parses uploaded spreadsheets, PDFs and photos more accurately and turns them into structured reports. It also pulls context from connected services such as Gmail and Google Docs. If you ask it to combine your correspondence with a supplier and a price table, it finds the emails and documents itself, so you do not need to summarize them by hand.
What are the key details on Tasks: more assignments within the same limits?
Tasks are multi-step jobs that the model carries out in the background while you do something else. According to Google, they run faster and more economically on Neon, so within the same Pro or Ultra subscription limits you should be able to complete more assignments. The company gives no specific speedup figures, so it is not possible to claim a difference of several times. For a product manager, this is a reason to move routine work into Tasks: a weekly competitor summary or the review of incoming documents.
What are the key details on Technical details and pricing for developers?
Neon's context window is 1M tokens. Standard output is capped at 64K tokens, and Long Decode mode raises it to 1M. There are four reasoning depth levels, so you can choose between a quick answer and a more thorough analysis. The deeper the model thinks, the longer and more expensive the response. In the API and in Antigravity, the model costs $3 per million input tokens and $15 per million output tokens. Cached input costs $0.15 per million, which is twenty times cheaper than regular input.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.