Maxwell is a new kind of frontier model from doubleAI that it calls a Code Execution Model. It sits between code and hardware, making programs run faster while they execute, without anyone changing a single line of code and without humans in the loop.
Unlike conventional tools, Maxwell sees programs from the inside, using traces, profiling information, and the data flowing through them. It adapts dynamically to the environment, including the hardware, other running workloads, and the shape of the inputs. The company argues that today's approach, where tools like Claude or Codex and human engineers repeatedly re-optimize code for each hardware and data scenario, is unnecessary, suboptimal, and does not scale.
Maxwell watches a program as it runs, rewrites the binary to fit that exact machine and data, and formally proves each improvement equivalent to the original code. Once it has enough runtime data, it swaps in these optimizations while the program keeps running. It works on both CPUs and GPUs.
The company reports 38x faster data processing pipelines and 4.25x faster AI inference kernels. In one featured case, Maxwell made LTX-2.5's 4K video generation nearly twice as fast, and in a Whisper speech-to-text test on an RTX 4090 it cut runtime from 106.23 seconds to 49.82 seconds. Early access is open at heymaxwell.ai.
Source: Hacker News · Summarized by HeadlinesBriefing