The open-source AI workbench for scientific research
One workspace for literature, code, experiments, compute, and results. Free models included, or bring Claude, GPT, Gemini and any other provider.
What is OpenScience?
OpenScience is an open-source agent that reads papers, writes code, and runs experiments with you.
- •Model agnosticFree models included, or your own keys for any provider
- •Scientific databasesUniProt, PDB, ChEMBL, PubChem, arXiv, and 37 more, as tools
- •Bundled skills371 skills across biology, chemistry, physics, ML, and writing, with a curated research core
- •ChatGPT Plus/ProSign in with OpenAI to use the subscription you already have
- •Manages computeBuilds environments and scales on demand: your laptop, cluster, or GPUs
- •Multi-sessionRun several agents in parallel on the same project
The state-of-the-art AI co-scientist
OpenScience leads every scientific benchmark we have run: Terminal-Bench Science, Terminal-Bench 4.0 (science), BiomniBench-DA, and OpenScience Bench, which measures end-to-end research. See the benchmark report.
Fig 1.75.7%Terminal-Bench Science
Fig 2.71.4%Terminal-Bench 4.0 (science)
Fig 3.82.2BiomniBench-DA
Fig 4.47.3%OpenScience Bench (pass@3)
Access reliable, optimized models for scientific agents
Ace gives you a handpicked set of models that OpenScience has tested and benchmarked for scientific agents, plus managed search and memory, behind one Wallet. No provider accounts, no inconsistent performance across routes.
Learn about Ace