ScienceRSI
A framework for recursive scientific self-improvement through specialization, verified capability transfer, and reuse across model scales.
55.0%4B SCI. REASONING AVG.
94.0%27B · 7-BENCHMARK AVG.
8B→27BCROSS-SCALE TRANSFER
SCINETICS / RESEARCH
Building intelligence
that can do science.
Starting with the life sciences, we develop science-native foundation models, autonomous scientific intelligence, and closed-loop experimentation.
FOUNDATION MODELS SCIENTIFIC REASONING AGENTS PROTEIN DESIGN AUTOMATED LABS
01 — LATEST RESEARCH
02 — FEATURED RESEARCH
Scientific Super Intelligence
From science tokens to ScienceRSI: a research agenda for models that reason directly across scientific modalities, recursively improve through verifiable evidence, and move toward Scientific Super Intelligence.
Science TokensSSI CrossoverScienceRSIEvidence-backed Scientific Assets
03 — PROTEIN DESIGN
Chain of Structures
Connecting reasoning and generation for protein design through explicit structural states—from catalytic cores and active sites to complete sequences and all-atom structures.
57.0%EC3 AGREEMENT
0.729TM-SCORE
4REASONING ROUNDS
04 — RESEARCH PROGRAM
One program, from representation to recursive learning.
01 / REPRESENTScience Tokens→
02 / REASONScientific Foundation Models→
03 / GENERATEChain of Structures→
04 / ACTSTELLA / BioClaw→
05 / EXPERIMENTLabOS→
06 / LEARNScienceRSI
05 — PREVIOUS WORK
Scientific intelligence, in motion.
Open research, models, and systems that established the foundations for what we are building today.
STELLA
Self-evolving AI Scientist
LabOS
Intelligent operating system for autonomous laboratories
PocketGen
Protein pocket generation and design
RNAGenesis
Generative RNA foundation model
BioMiner
Agentic scientific discovery system
ATOMICA
Multimodal learning for biomolecular systems
BUILD WITH US