Google Releases Gemini 4 Neon: Office Files Go Straight Into the API

Published on: 2026-11-19

Google Releases Gemini 4 Neon: Office Files Go Straight Into the API

On November 19, Google released Gemini 4 Neon, built for fast, end-to-end task completion across complex software engineering, knowledge work, and scientific workflows. It is the fourth Gemini 4 model in roughly seven weeks. The launch drew 37 reports from 26 sources that day and became the number-three trending story on AIHOT.

The most practical change was tucked inside the API update: you can now pass Word, PowerPoint, Excel, and other Office and OpenDocument files directly into the Gemini API, and split documents, tables, video, and audio within prompts and function responses. For ordinary users, that matters more than any benchmark. AI is moving from "type your question" toward "hand it your file."

One: reading Office files is the useful part

📖 Glossary

AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.

Before this, getting a document into an AI meant copying text, saving to another format, or extracting plain text yourself. The tedious part was never the analysis — it was the preparation.

Now the file goes straight in. Documents, spreadsheets, video, and audio can be split as needed. The reports, contracts, and slides sitting on your drive no longer need a human to carry them over first. It is not a flashy change, but anyone who works with files every day will feel it.

Two: the scorecard has two sources

KAIHE AIBOX concept device in a bright attic beside a READS YOUR FILES holographic panel

Read the results in two piles.

Third-party evaluations: Artificial Analysis scored its Intelligence Index at 61, one point above Carbon(high) and third overall; its coding-agent score was 59, on par with Claude Opus 5.5 and Claude Fable 5.2. In Arena's Agent Arena Index, Neon(High) took 1,336 points for second place, ahead of Carbon(High) at 1,303.

Google's own claims: first on the Coding Agent Index, 32% lower cost per task than Carbon, second in controllability, and the lowest fabrication rate in internal testing. Those are the vendor's words, not an independent verdict.

Put together, the fair reading is that Neon is strong on agentic coding tasks, but "outright first" overstates it.

Three: the catch — it actually costs more, and is harder to price

This is the part most likely to get lost in the spin.

The API unit price is higher: $3 per million input tokens, $15 per million output, and $0.15 per million cached input, with a one-million-token context window. Carbon and Argon are $2/$10; Hydrogen is $1.50/$7.50. Neon costs more than all of them.

So where does "32% cheaper" come from? That is Google's figure for cost per task, a different measure from unit price. Arena puts the median cost per task at $2.82, 53% below third-place Claude Opus 5.5(Max). The logic is that a faster, more token-efficient model can cost less per completed job even at a higher unit price.

Simon Willison, after testing it, said the launch makes Gemini 4's pricing more confusing. Four models in seven weeks, each with different prices and positioning — working out which to use takes some homework.

Four: three things worth keeping

KAIHE AIBOX concept device beside a TWO SCORECARDS holographic dashboard

First, ask who measured it. A third party's second or third place is not the same thing as a vendor's first.

Second, unit price and cost per task are separate ledgers. Neon costs more per token but may cost less per finished job. Do not blend the two.

Third, a file that can be fed in still needs somewhere to live. Uploading is easier; where the data goes and whether a machine has to stay on to process it are not solved by a new model.

Gemini 4 Neon is rolling out to Google AI Pro and Ultra subscribers in the Gemini App, the paid API tier, and business customers. It shortens the step of handing a file to an AI — for ordinary users, a natural companion is KAIHE AIBOX(铠盒AIBOX), a computer built specifically for agents and your local private AI assistant. It ships with OpenClaw and Hermes preinstalled, so it is ready once plugged in and connected, with no environment to rebuild. Inference runs in the cloud, while files, long-term memory, and call records stay on the local machine.

Models change faster than ever. Your files do not. The thing worth keeping is the one that holds them steady.

Recommended Products

A1 Home Entry A1 Pro Enhanced A2 Professional A2 Pro Advanced X1 Enterprise G1 Flagship
© KAIHE AI - Agent Computer Specialist