Skip to content
Neural Daily

ModelsNews

doubleAI's Maxwell speeds up code while it runs

doubleAI, the new company of former Mobileye CEO Amnon Shashua, has unveiled its first product. Maxwell is what doubleAI calls a code execution model: instead of writing code, it speeds up running programs and formally proves every change equivalent to the original.

By: Mikola Máté Patrik · · 2 min read

Maxwell by doubleAI: a blue cartoon creature holding a lantern next to the text The first Code Execution Model
Image: doubleAI · Press kit · Official site share image · Source

What a code execution model is

doubleAI argues that today's AI models help write code but never touch how it executes (doubleAI). Maxwell lives inside the running program. It watches the hardware and the data flowing through, finds bottlenecks and emits new binary code tailored to that exact machine, much like a just-in-time compiler. The source code stays unchanged.

The key is formal verification. Every replacement comes with a machine-checked proof that the new code gives the same result for the same input, with the same side effects. The company says it formalised the entire Python language to make this possible. Maxwell works on CPUs and GPUs, swapping Python for low-level, SIMD-optimised code and PyTorch code for kernels tuned to the GPU at hand.

What doubleAI showed

  • Whisper-small speech-to-text on an RTX 4090: 180 transcriptions in 49.82 seconds instead of 106.23, with identical output.
  • NetworkX betweenness centrality on CPU, 15 runs: 17.50 seconds instead of 61.44.
  • LTX-2.5 4K video generation on two B200s: 134 seconds per clip instead of 250, a 1.87x speedup.

Ctech reports that the company claims data pipelines more than 38 times faster and AI inference kernels more than 4.5 times faster on real-world workloads (Ctech). It adds that the figures have not been independently verified, and the announcement lacks detail on tests, hardware and baselines.

Funding and business model

According to Ctech, doubleAI raised a $200 million Series A co-led by Lightspeed Venture Partners and Bessemer Venture Partners, with Dell Technologies Capital, NVIDIA, Greenfield, Pitango and BRM participating. Ctech puts the valuation above $1 billion, with about 40 employees. The company plans to charge based on the compute savings it delivers.

Maxwell is available in early access via heymaxwell.ai.

Why it matters

Coding agents speed up writing code. Maxwell targets what happens next: how that code runs on the hardware. doubleAI pitches the two as complementary, with agents like Claude Code writing the program and Maxwell making the machine run it optimally. If formally proven optimisation scales, it means direct savings in a compute-constrained market. Until then, measure on your own workloads before planning around the 38x figure.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles