Google Launches Gemini, Its Native Multimodal Model to Rival GPT-4

Google today unveiled Gemini 1.0, the AI model the company had spent months teasing as its direct answer to OpenAI's GPT-4. Sundar Pichai, CEO of Google and Alphabet, and Demis Hassabis, who leads Google DeepMind, called it "our most capable and general model yet," according to the company's official announcement.

The news isn't just about performance. Google is emphasizing that Gemini is natively multimodal: unlike other systems that bolt image, audio or video modules onto a text model, Gemini was trained from the ground up to process and reason across those formats together. It's a technical distinction the company views as central to its results.

Three sizes for three audiences

Gemini comes in three versions, each built for a different scenario:

• Gemini Ultra, the most powerful model, built for complex tasks.
• Gemini Pro, designed to scale across a wide range of tasks.
• Gemini Nano, built to run directly on mobile devices without relying on the cloud.

Google has already moved on Pro: starting today, it powers the English-language version of Bard, the company's conversational assistant, replacing the model Bard ran on until now. Nano, meanwhile, is already running on the Pixel 8 Pro, enabling on-device features without sending data to external servers. Ultra, the flagship model, isn't available yet — Google says it will finish safety and trust testing before opening it up to developers and large customers in early 2024, when a more advanced version of Bard, called Bard Advanced, will also launch.

The GPT-4 comparison

The figure Google has pushed hardest is that Gemini Ultra beats GPT-4 on 30 of 32 standard academic benchmarks used to evaluate language models. The company singles out its result on MMLU (Massive Multitask Language Understanding, a test spanning 57 subjects from mathematics to professional ethics): Gemini Ultra reportedly scored 90%, surpassing human-expert-level performance on that test for the first time, according to Google.

It's the first time a lab has published numbers that directly challenge OpenAI's flagship model on its own benchmark turf — something that hasn't happened since GPT-4 launched in March. Until now, Google had answered ChatGPT with Bard, built on PaLM 2, without closing the perceived quality gap with GPT-4. Gemini represents the company's most ambitious bet yet to regain the initiative in a race where OpenAI and Microsoft have set the pace for months.

A demo that hasn't convinced everyone

Alongside the announcement, Google released a video titled "Hands-on with Gemini," in which the model appeared to respond in real time to images, drawings and objects shown to it, with a fluidity reminiscent of the multimodal demos OpenAI and others had shown months earlier only as conceptual mock-ups.

The video quickly drew skepticism. Several observers pointed out that the demo doesn't reflect a real-time interaction at all, but rather an edit built from still images and carefully selected text prompts — a far cry from the smooth conversational experience the video suggests. Google hasn't denied that the video was edited, but maintains that the capabilities shown are real, even if presented more briskly than everyday interaction would actually look.

What changes from here

With Gemini, Google is putting three pieces on the board that compete on different fronts: Nano against the on-device models Apple and other manufacturers want to build in, Pro against ChatGPT for everyday assistant use, and Ultra against GPT-4 in the fight to be the reference model for developers and enterprises.

The real test comes once Ultra is available outside Google's own controlled benchmarks and independent users can compare its performance against GPT-4 in real-world conditions. Until then, today's figures are, above all, a statement of intent: Google wants it clear that it is no longer trailing in the race for the biggest language models.