autoresearch vs AI-Scientist-v2
AI-driven machine learning research compared
autoresearch automates repeated language-model training experiments in a narrow, fixed-budget setup. AI-Scientist-v2 explores broader machine-learning research questions through generated ideas, experiments, and paper drafts, with less predictable outcomes.
autoresearch: Let AI Agents Run LLM Training Experiments
autoresearch gives an AI coding agent a small, single-GPU language-model training setup to modify and test autonomously. It suits researchers with supported NVIDIA hardware who want to explore training changes through repeated, time-bounded experiments.
AI-Scientist-v2: Automate Scientific Experiments and Paper Drafts
AI-Scientist-v2 uses LLM agents and tree search to propose research ideas, run experiments, analyze results, and draft papers. It is aimed at researchers exploring open-ended machine learning questions who can provide model APIs and a controlled compute environment.
| autoresearch | AI-Scientist-v2 | |
|---|---|---|
| Language | Python | Python |
| License | — | NOASSERTION |
| Stars | 97.2k | 7.3k |
| Forks | 13.5k | 995 |
| Last analyzed | Oct 3, 2026 | Oct 3, 2026 |
Key differences
- autoresearch has an agent modify a single training file and compare five-minute runs; AI-Scientist-v2 uses tree search to generate and evaluate experiment alternatives.
- autoresearch focuses on validation bits per byte for language-model training; AI-Scientist-v2 can search a user-described topic and produce a manuscript draft.
- autoresearch's main setup requires one NVIDIA GPU; AI-Scientist-v2 targets Linux with NVIDIA GPUs, CUDA, and PyTorch, and also requires access to supported model APIs.
- autoresearch provides Python code and reports 97.2k stars and 13.5k forks; AI-Scientist-v2 provides Python code and reports 7.3k stars and 995 forks.
- AI-Scientist-v2 reports a NOASSERTION license in its metadata and describes separate project-specific license terms; autoresearch's provided project details do not specify a license.
Choose autoresearch if you…
- want an AI agent to test language-model training changes under a consistent, five-minute budget.
- have a single NVIDIA GPU and want a compact setup focused on training experiments.
- want to inspect or adapt a small training setup for learning or experimentation.
Choose AI-Scientist-v2 if you…
- want to explore open-ended machine-learning questions through generated hypotheses and experiment paths.
- can provide supported model API access and a controlled compute environment for running generated code.
- want a pipeline that analyzes experiment results and produces a paper draft for human review.
Compare the numbers side by side, or add more projects
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.