A growing chorus of developers and researchers is pressuring Google to release Imagen (2022), Gemini 1.0 Nano, and Gemini 1.0 Pro as open-source models, arguing the company has little commercial reason left to keep them locked up.
It started, as these things often do, on Reddit and X. By April 16, a specific plea had worked its way to the top of several major technology forums: Google, please just open source Imagen, Gemini 1.0 Nano, and Gemini 1.0 Pro. You have nothing to lose at this point. It's a blunt ask, but the reasoning behind it is hard to dismiss.
The three models in question represent distinct chapters in Google's AI history. Imagen arrived in mid-2022 as a genuinely impressive text-to-image diffusion system, capable of photorealistic renders that had researchers genuinely excited. Google never released it publicly, citing safety concerns and, less explicitly, competitive positioning. Gemini 1.0 Nano and Pro came later, in December 2023, as the opening act of what Google positioned as its multimodal era. Both remain locked behind API paywalls and enterprise licensing agreements even as Google prepares to tease Gemini 3.0 for the second half of 2026.
The frustration is rooted in a simple observation: the competitive calculus that once justified keeping these models proprietary no longer holds. Meta's Llama series and the continued evolution of Stable Diffusion weights effectively democratized access to capable large language and image generation models through 2024 and 2025. What was a moat is now a memory. Gemini 2.0 Flash is currently the standard-bearer for efficiency and multimodal performance inside Google's lineup, which means the 1.0 series isn't protecting revenue so much as collecting dust behind a paywall.
There's a reputational dimension here that Google may be underestimating. The company has long cultivated goodwill in academic and research circles, but that reservoir has been draining as Meta, Mistral, and a range of open-source consortia have moved aggressively to publish model weights and encourage local fine-tuning. Researchers who want to study model behavior, run experiments without cloud infrastructure costs, or build educational tools are increasingly doing so on Meta's stack rather than Google's, simply because the weights are available.
Releasing Imagen 2022 and the Gemini 1.0 series would cost Google almost nothing in commercial terms while potentially recovering significant goodwill among exactly the community that shapes long-term platform loyalty. Developers who fine-tune on open weights tend to stay in the ecosystem. They write papers that cite the architecture. They build tools that extend the original work. That's free R&D influence, and right now Google is voluntarily forgoing it.
The Imagen case is particularly pointed. The model was celebrated on release for fidelity that surpassed contemporaries, yet it was never made accessible to the public in any meaningful form. Four years later, open-source image generation has caught up and in some respects moved ahead. The argument that releasing those weights would create safety risks or competitive harm grows weaker by the month.
What Google's silence actually signals
Google's reluctance to release legacy models is less about those specific models and more about the strategic posture the company has adopted as it competes directly with OpenAI and Anthropic for enterprise contracts. The logic runs that releasing older weights normalizes open-source access in ways that could eventually create pressure to release newer ones, blurring the line between Google's closed frontier models and the open ecosystem it has declined to fully embrace.
That's a defensible internal argument, but it carries a cost. The developer community reads the silence not as strategic discipline but as institutional hoarding, and that perception compounds with every passing quarter that models like Gemini 1.0 Pro sit unused and inaccessible while their technical relevance fades.
The broader question this movement raises is whether Google will eventually adopt a tiered release strategy similar to what Meta has practiced, where models beyond a certain age or capability threshold are routinely open-sourced as a matter of policy. That approach has served Meta's reputation in research circles extraordinarily well. For Google, the window to make that move with maximum goodwill attached is narrowing. Releasing Imagen, Gemini 1.0 Nano, and Gemini 1.0 Pro now would be a low-risk, high-signal gesture. Waiting until Gemini 2.0 is itself obsolete will make the same move feel obligatory rather than generous.
Also read: GPT-Rosalind launches as the first large language model built from the ground up for life sciences research • Anthropic's Claude Opus 4.7 benchmarks confirm the company's most capable model yet and pile pressure on OpenAI and Google • America's utilities are spending $1.4 trillion to keep the AI boom plugged in and homeowners will pay the difference
