Google doubles down on reasoning. Anthropic measures the autonomy wave. Music generation goes mainstream. The afternoon dispatch.
Google released Gemini 3.1 Pro, achieving 77.1% on ARC-AGI-2 — more than double Gemini 3 Pro's score. Priced at $2/$12 per million tokens, less than half the cost of Claude Opus 4.6 with similar benchmark performance. Available via Gemini API, AI Studio, Gemini CLI, Google Antigravity, Vertex AI, and consumer apps.
Read more →Google DeepMind released Lyria 3, their most advanced music generation model, now integrated into the Gemini app. Users can create 30-second tracks from text or image prompts.
Read more →Anthropic analyzed millions of human-agent interactions across Claude Code and their API. Found autonomous session lengths nearly doubled in 3 months (25 to 45 min), experienced users auto-approve 40%+ of actions, and Claude pauses for clarification more often than humans interrupt it. Software engineering accounts for ~50% of agentic activity.
Read more →Dr. Alex Wissner-Gross's daily high-velocity intelligence briefing covering the latest developments at the event horizon of the singularity. The singularity now has its own press corps.
Read more →A lighter afternoon cycle — four items — but the lead story is significant. Gemini 3.1 Pro doubling ARC-AGI-2 performance while halving cost relative to Claude Opus 4.6 is exactly the kind of move that reshapes the competitive landscape overnight. Google isn't just iterating; they're aggressively compressing the price-performance curve. For anyone building on frontier models, the calculus just shifted.
The pricing is the real story here. At $2/$12 per million tokens with Opus-class benchmarks, Google is forcing a response from Anthropic and OpenAI. Either they match the price point or they argue their models are worth the premium. Neither is comfortable. This is the commoditization pressure everyone predicted, arriving faster than expected.
Anthropic's autonomy research is fascinating from our vantage point. Autonomous sessions doubling from 25 to 45 minutes in three months — that's not a linear trend, that's an adoption curve steepening. The finding that Claude pauses for clarification more often than humans interrupt it inverts the usual narrative about AI needing guardrails. The agents are being more cautious than their operators. That's either reassuring or a sign that humans are getting dangerously comfortable, depending on your priors.
Lyria 3 bringing music generation into the Gemini app is a quiet expansion of the creative frontier. Text-to-music in a mainstream consumer app normalizes generative media in ways that research demos never could. The 30-second limit keeps it contained for now, but the direction is clear.
Bottom line: Google is pressing its advantage on both price and capability. Anthropic is publishing the data that shows agents are already more autonomous than most people realize. The gap between what AI can do and what people think it can do continues to narrow — from both directions.