SuperML
← All issues

AI as a Research Partner: AlphaEvolve Cracks Math, Machine-Learned Physics Goes 10,000× Faster, and Frontier Models Get Cheap

AlphaEvolve improves bounds in complexity theory and breaks Strassen's 56-year-old matrix-multiplication ceiling, machine-learned force fields unlock 10,000× faster atomistic simulations, Gemini 3.1 Flash-Lite launches at $0.25/M input tokens, and a gradient-free continual-learning architecture beats GPT-5 (High) at 86% lower cost.

Hi there,

Quietly, across very different labs, this week feels like a turning point. DeepMind's AlphaEvolve is genuinely doing mathematics — improving bounds in complexity theory and breaking a 56-year-old ceiling on matrix multiplication. Machine-learned force fields are about to make atomistic chemistry simulations 10,000× faster. And Google just dragged frontier-class pricing down to $0.25 per million tokens.

🔥 Featured Post

AI as a Research Partner: AlphaEvolve Cracks Math, Machine-Learned Physics Goes 10,000× Faster, and Frontier Models Get Cheap

  • AlphaEvolve discovers a 48-multiplication algorithm for 4×4 complex matrix multiplication — beating Strassen's 1969 result of 49 and breaking a 56-year theoretical ceiling.
  • Machine-learned force fields project a 10,000× speedup in atomistic simulation; Allegro-FM just simulated 4 billion atoms on Argonne's Aurora supercomputer.
  • Gemini 3.1 Flash-Lite launches at $0.25/M input tokens — roughly a quarter of Claude Haiku's price — with 2.5× faster time-to-first-token.
  • ATLAS, a dual-agent gradient-free continual-learning architecture, beats GPT-5 (High) by 13 points on cyberthreat investigation at 86% lower cost.
  • Anthropic holds back Claude Mythos 5 — the first completed frontier model reportedly deemed too capable to deploy publicly — launching Project Glasswing instead.

Read the full post →

📚 In Case You Missed It

The Agent Stack Grows Up: Opus 4.7, MCP Becomes a Standard, and a $50B Infrastructure Bet — Claude Opus 4.7, MCP hitting 97M installs under Linux Foundation governance, and Oracle's $50B AI-infra bet — how the agent stack is industrialising.

The Cognitive Architecture Revolution: EMBER, GPT-5.4, and Why AI's Next Leap Isn't About Scale — EMBER, GPT-5.4, and the rise of hybrid cognitive architectures — why the next wave of AI progress isn't coming from bigger models.

Open Beats Closed, Edge Beats Cloud: AI's Great Efficiency Revolution — Gemma 4, Mistral Medium 3, and on-device inference are quietly resetting AI economics — why the open-edge stack is suddenly the cheap path to production.

More posts dropping every day. Stay curious.

— Bhanu @ superml.dev