New Gemini 2.5 models' performance on WebDev and LMArena leaderboards

May 22, 2025

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> Google releases updates to Gemini 2.5 models; Pro model tops leaderboards

> announces “Deep Think” advanced reasoning mode - beats oAI’s models on USA Math Olympiad, LiveCodeBench and MMMU (multimodal reasoning)

> new Flash model beats o3 to take #2 on LMArena while being 67x cheaper and 11x cheaper in thinking mode

> announced $250/month "Ultra" tier with all Google AI features

> can now output audio, images and video + includes previews to text-to-speech mode

> $6.5B stock deal for ex-Apple Chief Designer Jony Ive's hardware startup io in OpenAI's biggest acquisition

> to take over all of OpenAI design; 55 expert engineers and researchers also set to join

> to develop AI devices described as "coolest technology the world will have ever seen" and "never happened before in consumer hardware" for AGI era

> market to reach $5 trillion by 2050 according to Goldman Sachs

> $2.3B projected this year, with companies in talks to raise:

  • Figure AI: $1.5B at $39.5B valuation

  • Skild AI: $500M at $4B valuation

  • Agility Robotics: $400M at $1.75B valuation

  • Beyond Imagination (by Ray Kurzweil): $100M at $500M valuation

  • Apptronik: already raised $403M at $1.5B valuation

> humanoid companies market share also dominated by US (32%) and China (27%)