Google is preparing Gemini 4: According to the company's CEO, it will be a real blast

An announcement worth noting was hidden in the last sentence of another announcement. Google introduced new Gemini Flash series models on July 21st and added a note at the end of the text that it had already launched its most ambitious pre-training to date – for Gemini 4. Two days later, Sundar Pichai elaborated on this to analysts.

What Pichai told analysts

At Wednesday’s conference on Alphabet’s second-quarter results, the company’s CEO described Gemini 4 as a significantly larger model than Google has ever trained, and said he was pleased with the internal progress. However, one of his formulations is key: Google is reportedly building the model to perform at the cutting edge of the industry at the time of its release – not at today’s. In other words, the company is estimating where the competition will be in a few months and aiming for that.

When a Barclays analyst asked him how Google plans to keep pace with rivals who release top-tier models much more frequently, Pichai replied that the company is building a foundation upon which it can iterate quickly. It aims to approach an almost monthly release cadence.

Yet, version Three still hasn’t been released

This is where things start to get complicated. Gemini 3.5 Pro was already showcased by Google at the I/O conference, with a general release planned for June. To this day, it remains limited to partner testing as the company addresses reliability and performance in programming. So, Google is publicly talking about version four while not having delivered 3.5 – which is either confidence or an attempt to redirect attention. Probably both.

Pichai also unusually openly admitted where Google is lagging: in programming and so-called agentic coding, meaning the model’s ability to autonomously work on code in several steps. He said something similar earlier in the Hard Fork podcast, where he explained the deficit by stating that Google lacked a developer tool through which it could collect the same data as the competition.

What we don’t know about Gemini 4 yet

Practically everything. Google has not disclosed the number of parameters, architecture, context window size, prices, test results, or release date. Only the ongoing training itself and two general statements from the CEO have been confirmed. Colleagues from 9to5Google estimate November or December based on previous releases, but this is an estimate, not information.

Nothing was said about language support either, so we don’t know yet if the new capabilities will be available in Czech from day one. With Gemini, it usually turns out well, but promising it based on a single sentence in a press release would be premature.

Another piece of information from the same conference is also noteworthy: Alphabet increased its estimate for this year’s infrastructure investments to up to 205 billion dollars. The company’s shares fell by more than three percent after the results were published. The market is therefore not yet convinced that this money is yielding returns.

Do you use Gemini daily, or do you prefer to stick with the competition?

Sources: 9to5Google, Benzinga, iTechPost