Gemini API cuts costs 75% + Alibaba releases Qwen3 LLMs

Here’s a categorized summary of the key AI product management discussions and announcements:

Major Product Launches & Updates

• Logan K announced significant updates to Google AI Studio, noting it will be “barely scratching the surface” of its capabilities by year-end
• Google launched implicit caching in Gemini API, enabling 75% cost savings with Gemini 2.5 models when requests hit cache
• OpenAI rolled out memory improvements for Plus and Pro users in EEA, UK, and other European regions
• Alibaba released Qwen3, featuring eight open LLMs including two mixture-of-experts models, supporting 119 languages

Product Strategy & Management Insights

• Andrew Ng shared insights on AI Fund’s $190M raise and emphasized speed as crucial for startup success
• Shreyas noted that purely metrics-driven product decisions might be the first to get automated by AI
• Claire Vo observed that production releases can now be treated like disposable prototypes due to AI capabilities

AI Applications & Integration

• Netflix is testing generative AI search capabilities powered by OpenAI models
• Figma released comprehensive AI tools including Figma Make, Sites, Draw, and Buzz
• Microsoft announced adoption of Google’s Agent2Agent protocol for Azure AI Foundry and Copilot Studio

Research & Technical Developments

• Meta introduced Meta Locate 3D for accurate object localization in 3D environments
• Google DeepMind celebrated AlphaFold 3’s first anniversary, highlighting its impact on molecular research

Memes & Humor

• Claire Vo shared a humorous take on meeting durations: “10 min ad hoc call or 1 hour+ jam or 3 days in person nothing in between”