alphaXiv puts an army of agents to work reproducing arXiv papers, with Tinker paying the compute bill
The new autoresearch feature hands a paper to Claude and Codex agents and lets them run fresh experiments on it. Training runs go through Thinking Machines' Tinker API.
alphaXiv, a platform for reading research papers, has announced a feature it calls autoresearch. The idea: you open an arXiv paper and the platform sets a group of agents to reproducing it. The agents run on Claude and Codex, writing the code, setting up the experiments and reporting results.
The other half of the problem is compute. For agents to launch training and reinforcement learning runs, those runs have to happen somewhere, and alphaXiv has made Thinking Machines' Tinker API the default backend. The platform's compute screen also offers the user's own machine, a host reached over SSH, Hugging Face Jobs, Modal and Slurm.
Experiments do not proceed in a single line. Each attempt branches, growing an experiment tree the user can watch expand. Tinker describes it as turning papers from static artifacts into living research that grows and branches.
The feature is available in alphaXiv's OpenResearch section.
