GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
← Back to feed

AlphaEvolve: DeepMind's Gemini Coding Agent Hits 88% on Power Grids, 30% Fewer DNA Errors

Google DeepMind published a one-year impact update for AlphaEvolve today, detailing deployments across scientific and industrial domains. The numbers are specific.

Power Grid Optimization

AlphaEvolve was applied to the AC Optimal Power Flow problem — a core challenge in electricity grid management. The system raised the share of feasible solutions found by a trained Graph Neural Network from 14% to over 88%, sharply reducing the need for costly post-processing steps.

Genomics: 30% Fewer Sequencing Errors

Working with PacBio, AlphaEvolve improved DeepConsensus, Google Research’s DNA sequencing error-correction model. Variant detection errors dropped 30%. PacBio’s Senior Director Aaron Wenger: “The solution the Google team discovered using AlphaEvolve unlocks meaningfully higher accuracy rates for our sequencing instruments.”

Quantum Computing: 10x Lower Circuit Error

AlphaEvolve suggested quantum circuit optimizations that reduced error rates by 10x compared to conventionally optimized baselines on Google’s Willow quantum processor. The improvements enabled first-of-a-kind experimental demonstrations.

Mathematics

AlphaEvolve improved lower bounds on the Traveling Salesman Problem and Ramsey Numbers — two classic open challenges in combinatorics. Working with Terence Tao, it contributed to solving Erdős problems. Tao: “Tools such as AlphaEvolve are giving mathematicians very useful new capabilities.”

Infrastructure

AlphaEvolve is already deployed inside Google’s own computing infrastructure, optimizing algorithms used at production scale. DeepMind has not disclosed which specific components.

What It Is

AlphaEvolve uses Gemini as the generative core of an evolutionary search loop. It proposes, scores, and iteratively improves algorithms against objective evaluators — no human in the loop for the refinement cycles. The system was first announced in May 2025.

The breadth of the impact update — health, energy, physics, mathematics, cryptography, safety — signals that DeepMind is positioning AlphaEvolve as a general-purpose scientific accelerant, not just a coding benchmark entry.