Schrödinger introduced Bunsen on 27 July, an agentic system it describes as an AI co-scientist for molecular discovery. Early access is open to selected customers now, with full commercial release expected by the end of 2026.
What it does
Bunsen runs a four-stage loop — Understand, Plan, Execute, Interpret — orchestrating Schrödinger's existing physics-based computational chemistry tools rather than replacing them. It is built on NVIDIA's BioNeMo Agent Toolkit, runs on RTX PRO 6000 Blackwell Server Edition GPUs with Google Cloud infrastructure, and is named after Robert Bunsen. Chief executive Ramy Farid says it "removes friction from the discovery cycle."
The evidence section is empty
This is the part worth stating plainly. The announcement carries no hit rate, no discovery success metric, no head-to-head comparison against human-directed workflows, and no validation study. For a product whose entire claim is that it accelerates scientific discovery, the absence of a single quantitative result is not a minor omission — it is the whole question, unanswered. "AI co-scientist" is branding until somebody publishes a number.
What the AI is actually doing
Not finding molecules. Schrödinger's differentiator has always been physics-based simulation — free energy perturbation and related methods that compute binding affinity from first principles. Bunsen is a planning and interpretation layer above those engines. The discovery work is still being done by the physics; the agent decides what to run and reads the output. That is a genuinely useful thing to automate, and it is a much narrower claim than the branding implies.
Why it is notable anyway
Because of who is conceding what. Schrödinger built its business arguing that simulation beats machine learning for molecular prediction. Wrapping that stack in an agent is an acknowledgement that the interface to computational chemistry — deciding which calculations to run, in what order, and what the results mean — is now the bottleneck rather than the calculations themselves.