**Sonnet 5 & Step 3.5 Flash Set to Disrupt AI as DeepMind’s Aletheia Cracks Math Milestone**
New AI Models & Releases
Sonnet 5 Expected Next Week: Rumored to launch next week, Sonnet 5 is anticipated to feature a 1M-token context window, cost half as much as Opus 4.5, and outperform it across all benchmarks. The model is trained on TPUs and is expected to excel in agentic coding tasks.
Step 3.5 Flash (196B/11B) Outperforms GLM-4.7 and DeepSeek v3.2: StepFun AI’s new model, Step 3.5 Flash, boasts 196B total parameters (11B active) and surpasses competitors in coding and agentic benchmarks despite its smaller active parameter count.
Mistral Vibe 2.0 Released: Mistral AI’s updated model introduces improvements and new features, continuing its push for advanced AI capabilities.
AI Research & Breakthroughs
DeepMind’s Aletheia Solves Erdős-1051 Problem Autonomously: The new reasoning agent, powered by Gemini Deep Think, autonomously solved the Erdős-1051 math problem, marking a milestone in AI-driven mathematical research.
AI Tools & Platforms
AI-Powered Pentesting Platform with 400+ Hacking Tools: A new open-source platform enables AI agents to execute pentesting tools in Docker, chain attacks, and auto-document findings, streamlining security testing.
ACE-Step 1.5: Open-Source Music Generation on <4GB VRAM: A lightweight, local alternative to Suno v4.5/v5, ACE-Step 1.5 enables music generation without subscriptions or API limits, launching in one day.
Business & Monetization
OpenAI Launches Beta Ads on ChatGPT with $200K Minimum Spend: OpenAI is testing ads in ChatGPT, requiring selected advertisers to commit at least $200,000, signaling a push toward ad-based monetization.