A year after DeepMind shipped AlphaEvolve, they put up the scorecard. 32.5% speedup on FlashAttention kernels. 4×4 complex matrix multiplication in 48 scalar multiplications — first time anyone beat Strassen since 1969. 0.7% of Google’s worldwide compute recovered daily. 10x lower error on Willow quantum circuits. Not a demo.
What it actually is
An evolutionary coding agent, not a chatbot. Gemini Flash generates wide code mutations, Gemini Pro reasons about which to keep, an evaluator scores each on a real metric — runtime, error rate, compute saved. Survivors breed the next generation. It evolves whole codebases, not single functions. Give it any problem with measurable feedback and it iterates for hours on real engineering targets, not benchmarks.
Why the report card matters
Spanner write amplification down 20%. Circuits going into next-gen TPU silicon. DNA variant detection errors cut 30% at PacBio. Schrödinger’s force-field training 4x faster. Klarna doubled transformer training throughput. FM Logistic shaved 15,000 km off yearly routes.
Google Cloud is opening AlphaEvolve to enterprise customers in private preview. Bring a code-editable problem with a measurable score; AlphaEvolve evolves the solution. First time industrial-strength algorithm discovery sits behind a commercial API. Every shop with a slow inner loop and a benchmark just got a new option.
You Might Also Like
- Alphaevolve Year one 32 5 Faster Flashattention a Fallen Strassen Record now on Google Cloud
- Cloudrouter Gives Your ai Coding Agent its own Cloud Machine and Thats a big Deal
- Google Deepmind Aletheia Just Solved Math Problems Nobody Could Heres why That Matters
- Google A2ui Agent to User Interface Finally a Standard way for ai Agents to Show you Things
- Google Lyria 3 Just Turned Gemini Into a Music Studio and im Weirdly Into it

Leave a comment