Just two months after its initial release of the Gemini AI models, and one week after the release of Gemini 1.0 Ultra, its most capable AI model, Google today released its first Gemini 1.5 model, Gemini 1.5 Pro. And it’s bringing Gemini 1.0 Ultra to developers for the first time as well.
Confused? Buckle up and join the club. After a year of talking about AI, Google has finally stepped on the gas pedal.
Sign up for our new free newsletter to get three time-saving tips each Friday — and get free copies of Paul Thurrott's Windows 11 and Windows 10 Field Guides (normally $9.99) as a special welcome gift!
"*" indicates required fields
“Gemini 1.5 shows dramatic improvements across a number of dimensions and 1.5 Pro achieves comparable quality to 1.0 Ultra, while using less compute,” Google CEO Sundar Pichai writes in the announcement post. “This new generation also delivers a breakthrough in long-context understanding. We’ve been able to significantly increase the amount of information our models can process — running up to 1 million tokens consistently, achieving the longest context window of any large-scale foundation model yet. Longer context windows enable entirely new capabilities and help developers build much more useful models and applications.”
“Gemini 1.5 delivers dramatically enhanced performance. It represents a step change in our approach, building upon research and engineering innovations across nearly every part of our foundation model development and infrastructure,” Google Deepmind CEO Demis Hassabis adds. “This includes making Gemini 1.5 more efficient to train and serve, with a new Mixture-of-Experts (MoE) architecture.”
Today, Google is releasing only the Gemini 1.5 Pro model, but I assume we will see Ultra and Nano variants sometime soon as well. Hassabis describes Gemini 1.5 Pro as a mid-size multimodal model that’s optimized for scaling across a wide range of tasks. That it performs similarly to the much bigger Gemini 1.0 Ultra should be a wake-up call to Google’s competitors at OpenAI and Microsoft, if accurate.
For now, Gemini 1.5 Pro comes with a 128,000 token context window, but Google is opening up a 1 million context window to a limited group of developers and enterprise customers via AI Studio and Vertex AI in private preview. This capability will roll out to more customers after it gets feedback and performs further optimizations to improve latency, reduce computational requirements, and enhance the user experience, Hassabis says.
The big deal here appears to be the model’s “context window,” which Google explains is made up of tokens—full or partial words, code, images, audio, or videos—the building blocks that AI models use to process information. Where Gemini 1.0 supported 32,000 tokens, Gemini 1.5 is running up to 1 million tokens in production, which is powerful enough to process over 700,000 words, codebases with over 30,000 lines of code, 11 hours of audio, or 1 hour of video in a single pass. And the firm claims it has successfully tested up to 10 million tokens.
Developers interested in testing Gemini 1.5 Pro can sign up now in AI Studio, while enterprise customers are advised to reach out to their Vertex AI account team.