Gemini 4 Is Training: Google Confirms Its Most Ambitious Model Yet

Google has officially entered the Gemini 4 era — sort of. On July 21, 2026, alongside new Flash models, Google confirmed it had begun its most ambitious pretraining run yet for Gemini 4. Two days later, on the July 23 earnings call, Sundar Pichai described it to analysts as a "significantly larger" frontier model.

What's Actually Confirmed

Here's the important distinction for anyone reading breathless "Gemini 4 specs leaked" headlines:

• Confirmed: Pretraining has started. Gemini 4 will be significantly larger than the current generation.
• Not confirmed: Release date, model ID, price, context window, and benchmarks. None of these have been published.

The widely repeated "November or December 2026" window is analyst inference from Google's historical release spacing — not a company commitment. Treat leak roundups accordingly.

Meanwhile: Gemini 3.6 Flash Shipped

Google isn't standing still while Gemini 4 trains. On the same day, it launched Gemini 3.6 Flash — a 1M-context workhorse priced at $1.50 / $7.50 per 1M tokens, faster and cheaper than 3.5 Flash. There's also a 3.5 Flash-Lite tier for high-volume, low-cost tasks.

The Gemini 3.5 Pro Saga

The subplot here is Gemini 3.5 Pro, which Pichai promised at Google I/O (May 19, 2026) would arrive "within a month" — and still hadn't shipped by late July. Bloomberg reported the delay was tied to coding performance falling short of internal targets, with a late-June training-data update reportedly making results worse. Google appears to have pivoted energy toward Gemini 4 instead.

The lesson: even for a lab with Google's resources, the frontier is hard. Delays and unglamorous "the coding scores aren't good enough" problems are now part of the story.

The Bottom Line

Gemini 4 is real, it's big, and it's training — but nearly everything else you'll read about it is speculation. The nearer-term action is in the Flash tier, where Google is competing aggressively on price and context length.

Brandomize tracks the frontier so your content strategy doesn't have to guess. Want a model-agnostic AI workflow that survives every new Gemini? Let's build it.