
Hard Fork
The AI community has been buzzing since Google unveiled Gemini 3, the latest iteration of its DeepMind‑backed language model. In internal briefings the company highlighted a jump from 21.6 % to 37.5 % on the demanding “Humanity’s Last Exam,” a benchmark that simulates graduate‑level interdisciplinary problem solving. Compared with Gemini 2.5 Pro, the new model consistently beats its predecessor across more than a dozen standard tests, positioning Google as a serious contender in the race that has been dominated by OpenAI and Anthropic. Analysts see this performance gap as a potential shift in market leadership, especially as competitors scramble to match the gains.
Beyond raw scores, Gemini 3 introduces several product‑level innovations. The model can synthesize custom interactive interfaces on the fly—building tutorials about Van Gogh or mortgage calculators without developer input. Google also demonstrated a “Gemini agent” that scans a user’s inbox, suggests replies, and organizes messages, tackling a long‑standing productivity wish‑list. Improvements in multi‑step reasoning, front‑end coding (vibe‑coding), and a more concise, pleasant persona were repeatedly emphasized by executives Demis Hassabis and Josh Woodward. These capabilities aim to make the model a true “super‑tool,” blending research assistance, software creation, and everyday task automation.
Google’s rollout strategy reflects its massive ecosystem advantage. Gemini 3 Pro is already available in the Gemini app and the AI‑mode side tab of Google Search, suggesting the model can be served at scale without prohibitive cost. Enterprise customers will soon access the model via Google Cloud, while billions‑user products like Gmail and Docs are slated for later integration. In a bold educational move, the company is granting U.S. college students a free year of Gemini Pro, sparking debate over academic integrity but also seeding future user adoption. With efficiency gains from model distillation, Google appears poised to leverage Gemini 3 as both a revenue driver and a long‑term foundation for its AI‑first product suite.
Google’s much anticipated new large language model Gemini 3 begins rolling out today. We’ll tell you what we learned from an early product briefing and bring you our conversation with Google executives Demis Hassabis and Josh Woodward, just ahead of the launch.
Guests:
Demis Hassabis, chief executive and co-founder of Google DeepMind
Josh Woodward, vice president of Google Labs and Google Gemini
Additional Reading:
The Man Who ‘A.G.I.-Pilled’ Google
We want to hear from you. Email us at hardfork@nytimes.com. Find “Hard Fork” on YouTube and TikTok.
Subscribe today at nytimes.com/podcasts or on Apple Podcasts and Spotify. You can also subscribe via your favorite podcast app here https://www.nytimes.com/activate-access/audio?source=podcatcher. For more podcasts and narrated articles, download The New York Times app at nytimes.com/app.
Comments
Want to join the conversation?
Loading comments...