Gemma Hits 900M Downloads, Giving Google a Real Claim in the Open-Weights Race

2. Gemma Hits 900M Downloads, Giving Google a Real Claim in the Open-Weights Race

Demis Hassabis posted on X on July 25, 2026, clarifying that the Gemma open model series has now crossed 900 million total downloads, with Gemma 4 alone accounting for 300 million of those. The post came in direct response to Jensen Huang's July 24 letter, co-signed by Nvidia, arguing that open models strengthen safety, accelerate innovation, and enable national AI sovereignty. Hassabis framed Gemma's numbers as evidence that Google DeepMind has "always supported and contributed heavily to open source," citing JAX, the Transformers library, and AlphaFold alongside Gemma.

The 900M figure matters because the open-weights competitive landscape has been largely defined by Meta's Llama series, which set the reference point for community adoption. Google has historically struggled to convert its research credibility into grassroots developer traction. A nine-figure download count changes that narrative. If Gemma 4 alone is at 300M, the series is compounding fast. That gives Google a concrete answer to the question of whether its open releases are symbolic gestures or genuine infrastructure choices developers depend on. It also arrives at a moment when the policy debate around open models is heating up: Huang's letter signals that Nvidia is actively lobbying on this front, and Google now has a usage number to put behind any similar position.

The next number to watch is whether Gemma's download velocity continues to outpace earlier generations. Download counts can be inflated by automated pulls and mirror traffic, so the more telling signal will be community fine-tune activity and third-party deployment reports. If Gemma 4 shows up in the same downstream toolchains where Llama 3 currently dominates, Google's claim to open-ecosystem leadership becomes structurally real rather than statistically convenient.

Source: @demishassabis on X