Oriol Vinyals
@OriolVinyalsML
VP of Research & Deep Learning Lead, Google DeepMind. Gemini co-lead. Past: AlphaStar, AlphaFold, AlphaCode, WaveNet, seq2seq, distillation, TF.
Ahead of I/O, we’re releasing an updated Gemini 2.5 Pro! It’s now #1 on WebDevArena leaderboard, breaking the 1400 ELO barrier! 🥇 Our most advanced coding model yet, with stronger performance on code transformation & editing. Excited to build drastic agents on top of this!
Cool benchmark! Good to see Gemini 2.5 topping another leaderboard, though we are quite far from the summit 😅🪜🏔️
Can GPT, Claude, and Gemini play video games like Zelda, Civ, and Doom II? 𝗩𝗶𝗱𝗲𝗼𝗚𝗮𝗺𝗲𝗕𝗲𝗻𝗰𝗵 evaluates VLMs on Game Boy & MS-DOS games given only raw screen input, just like how a human would play. The best model (Gemini) completes just 0.48% of the benchmark! 🧵👇
Drastic progress on maths with Gemini 2.5! As a math undergrad, I am impressed 🤯 🥈 -> 🥇 ✅ Formal -> Informal ✅ Specialized model -> General model ✅ Available soon ✅ Huge thanks to IMO and congrats to all participants! Blog: deepmind.google/discover/blog/…
"A Neural Conversational Model" is 10 years old, w/ @quocleix . TL;DR you can train a chatbot with a large neural network (~500M params!). Samples 👇 This paper was received with mixed reviews, but I'm glad all the critics are now riding the LLM wave 🌊 arxiv.org/abs/1506.05869
Hello Gemini 2.5 Flash-Lite! So fast, it codes *each screen* on the fly (Neural OS concept 👇). The frontier isn't always about large models and beating benchmarks. In this case, a super fast & good model can unlock drastic use cases. Read more: blog.google/products/gemin…
Introducing the new Gemini 2.5 Pro preview, with a +24 LMArena Elo score over its predecessor! 📈 It leads on tough coding (AIME, AIDER), science (GPQA), and reasoning (HLE). Style/structure have improved thanks to your feedback. Learn more: goo.gle/4kKynYo
Prompt Theory (Made with Veo 3) What if AI-generated characters refused to believe they were AI-generated?
As highlighted yesterday at #GoogleIO, Gemini 2.5 Pro with #DeepThink mode achieved a new state-of-the-art performance of 49.4% on USAMO 2025, an extremely challenging Math benchmark for AI that requires detailed proof writing. I was analyzing the model outputs with our math…
It's been a wild run over the last 2 weeks to land Deep Thinking just in time for I/O! Heroic efforts across many teams @GoogleDeepMind from research to production that felt almost impossible last month! Proud of our research team for the relentless effort over the past 6 months…
Today we introduced Gemini Diffusion⚡️ (& DeepThink, Veo3, Imagen4, 2.5 updates...). It's been a dream of mine to remove the need for "left to right" text generation. It's so fast, that we had to *slow down* the video during the presentation. deepmind.google/models/gemini-…
We may be delaying AGI if we keep playing this game, but OK! I used Copy Canvas and here is my addition to the Cartographer. Coding this in two sentences... drastic & wild times! @NoamShazeer I nominate you : ) g.co/gemini/share/c…
This is cool! You can click "Copy Canvas" in the lower right (after logging in) and enable voice activation.. (volume up for video below) @OriolVinyalsML tagging you! Thank you @davemesserx!