Google Deepmind has upgraded its specialized thinking mode "Gemini 3 Deep Think" and made it available through the Gemini app and as an API via a Vertex AI early access program. The upgrade targets complex tasks in science, research, and engineering.
Google AI Ultra subscribers can access Deep Think through the Gemini app, while developers and researchers can sign up separately for the API program. According to Google Deepmind, the model tops several major benchmarks:
Benchmark
Deep Think
Claude Opus 4.6
GPT-5.2
Gemini 3 Pro Preview
ARC-AGI-2 (Logical reasoning)
84.6%
68.8%
52.9%
31.1%
Humanity's Last Exam (Academic reasoning)
48.4%
40.0%
34.5%
37.5%
MMMU-Pro (Multimodal reasoning)
81.5%
73.9%
79.5%
81.0%
Codeforces (Coding/algorithms, Elo)
3,455
2,352
-
2,512
While Deep Think dominates in logic and coding, the gap narrows significantly on MMMU-Pro: it scored 81.5 percent, barely ahead of Gemini 3 Pro Preview at 81.0 percent. This suggests the thinking upgrades focus heavily on abstract reasoning rather than visual processing. Deep Think also achieved gold medal-level results at the 2025 Physics and Chemistry Olympiads. Examples of the model in scientific use can be found here.
These games provide objective ways to measure skills like planning and decision-making under uncertainty. Gemini 3 Pro and Gemini 3 Flash currently hold the top spots in all rankings. The Werewolf benchmark serves double duty for security research as well: it tests whether models can detect manipulation without any real-world consequences. According to Google Deepmind CEO Demis Hassabis, the AI industry needs more rigorous tests to properly evaluate the latest models.
Demis Hassabis (Google Deepmind) and Dario Amodei (Anthropic) are already seeing early signs of AI's impact on the job market. Speaking at the World Economic Forum, Hassabis said entry-level jobs and internships could take a hit this year, something he's already noticing at Deepmind. In the near term, new and potentially more meaningful jobs could emerge, Hassabis said, but once AGI (artificial general intelligence) arrives, we're in uncharted territory. He criticized governments and economists for failing to grasp the scale of the changes ahead.
Amodei is sticking with his prediction that half of office jobs for young professionals could vanish within one to five years. Like Hassabis, he says he's already seeing this at Anthropic, where the company expects to need fewer junior and mid-level employees going forward. AI could outperform humans at everything within one to two years, he says, but the labor market is slow to react. His concern: the exponential pace of development will outstrip our ability to adapt.
Deepmind co-founder Shane Legg puts the odds of achieving "minimal AGI" at 50 percent by 2028. In an interview with Hannah Fry, Legg lays out his framework for thinking about artificial general intelligence. He describes a scale running from minimal AGI through full AGI to artificial superintelligence (ASI). Minimal AGI means an artificial agent that can handle the cognitive tasks most humans typically perform. Full AGI covers the entire range of human cognition, including exceptional achievements like developing new scientific theories or composing symphonies.
Legg believes minimal AGI could arrive in roughly two years. Full AGI would follow three to six years later. To measure progress, he proposes a comprehensive test suite: if an AI system passes all typical human cognitive tasks, and human teams can't find any weak points even after months of searching with full access to every detail of the system, the goal has been reached.
Demis Hassabis, CEO of Google Deepmind, expects the next year to bring major progress in multimodal models, interactive video worlds, and more reliable AI agents. Speaking at the Axios AI+ Summit, Hassabis noted that Gemini's multimodal capabilities are already powering new applications. He used a scene from "Fight Club" to illustrate the point: instead of just describing the action, the AI interpreted a character removing a ring as a philosophical symbol of renouncing everyday life. Google's latest image model uses similar capabilities to precisely understand visual content, allowing it to generate complex outputs like infographics, something that wasn't previously possible.
Hassabis says AI agents will be "close" to handling complex tasks autonomously within a year, aligning with the timeline he predicted in May 2024. The goal is a universal assistant that works across devices to manage daily life. Deepmind is also developing "world models" like Genie 3, which generate interactive, explorable video spaces.
Isomorphic Labs, a Deepmind spin-off focused on drug discovery, is getting ready for its first clinical trials with drugs designed using AlphaFold-based AI models.
"We're staffing up now. We're getting very close," said Colin Murdoch, President of Isomorphic Labs and Chief Business Officer at Deepmind, in an interview with Fortune.
The company wants to overhaul the traditionally slow and expensive process of drug development, with anti-cancer drugs already in the pipeline. Isomorphic Labs has signed agreements with Eli Lilly and Novartis, and in 2025 closed a USD 600 million investment roundled by Thrive Capital.
Looking ahead, Isomorphic Labs has even bigger ambitions for AI in medicine. "One day we hope to be able to say— well, here's a disease, and then click a button and out pops the design for a drug to address that disease," Murdoch said.
