Get a quote

Google makes Gemini 4 Neon generally available via API

Google released Gemini 4 Neon to the paid tier of the Gemini API with direct Word, Excel and PowerPoint input. Here is what Gulf, Egyptian and Lebanese companies should check before sending it customer data.

Google made Gemini 4 Neon generally available in the Gemini API on November 18, at $3 per million input tokens and $15 per million output tokens. The release is listed in the Gemini API changelog, which describes the model as built for software engineering, knowledge work and science tasks.

Neon is on the paid tier only, with no free tier access, according to Google's developer guide for the model. Cached input costs $0.15 per million tokens, a 95% discount that applies automatically to repeated prompts of at least 4,096 tokens. Input, thinking and output share one window of 1,048,576 tokens, and developers choose between four thinking levels, low, medium, high and xhigh, with high as the default. The guide says xhigh adds latency and token use.

The same changelog entry adds direct input of Word, Excel and PowerPoint files, and lets developers send only selected pages, worksheet tabs or cell ranges, so they pay for that part alone.

What changed

Google places Neon between Gemini 4 Hydrogen, its everyday model, and Gemini 4 Carbon, its most capable reasoning model. For now the prices do not follow that order: the changelog lists Hydrogen at $1.50 and $7.50 per million tokens, and Carbon at an introductory $2 and $10 through November 30, 2026. Until that date, Carbon costs less per token than Neon. Google says Neon completes tasks with leaner token use, so the cost of a finished task can differ from what the list price suggests.

What Saudi and UAE companies should check before using Neon

Access is not the obstacle. Google's list of available regions for the Gemini API names Saudi Arabia, the United Arab Emirates, Egypt, Jordan and Lebanon.

Location is the question. The Gemini 4 Neon page on Google Cloud offers the model through a global endpoint and the US and EU multi-regions, and lists machine learning processing in the US or the EU. Google's endpoint locations page shows Doha and Dammam as Middle East locations, but the Neon page names neither. The same model page says Neon is offered to verified enterprise customers in select regions.

That matters when prompts carry customer data. Under SDAIA's regulation on transferring personal data outside the Kingdom, a controller relying on safeguards such as standard contractual clauses must complete a risk assessment before the transfer, as summarised by the law firm CMS. This likely means a short written assessment before a Saudi support bot or invoice workflow sends names, phone numbers or account details to Neon.

One default is easy to miss. Since October 26, new paid projects on the Generate Content API store requests by default, according to the changelog, and opting out takes a store=false setting or a change in the logging settings.

What an owner can do this week:

  1. Start with files that hold no personal data, such as supplier price lists or internal policies kept in Excel and Word.
  2. Set store to false on any request that carries customer details.
  3. Run the same task on Neon and on Carbon before November 30, while the Carbon introductory price lasts.

Voxire builds SaaS products and web apps for companies across the Middle East and can connect a model like Neon to the systems you already run: see the SaaS development service.

Sources

Voxire

SaaS for MENA

The Voxire service closest to this story. See what it covers, what it costs and how the team works.

Learn more

Does this change your plans for a website, a store or a product? Voxire builds and grows them for businesses in Lebanon and across MENA.

All news

More news

Chat on WhatsApp