C5R CORP introduces SciUniverse after building an AI-run research facility

In 12 weeks, we built a research facility that is run entirely by AI.

AI designs, executes, and observes experiments end-to-end across biology, chemistry, and materials science.

We’re introducing SciUniverse: a benchmark that measures AI’s ability to do real-world scientific research.

We made an ontology of actions, an inventory management system, and an experiment scheduler across machines & people. We’ve integrated 40+ scientific instruments – by reverse engineering drivers and using custom hardware adapters.

This is a physical harness for research.

Once given a goal, models can explore inventory and read specs to design experiments as code. The code is converted into equipment control and instructions to people.

Models can then analyze measurements, replan and make a choice about what to try next – just like a person.

But can they do science in the real world?

To answer this we made SciUniverse [Level 1], a scientific benchmark that spans the research process including protocol debugging and bring-up, instrument operation, data analysis, and longer horizon tasks.

We learned models have theoretical scientific knowledge, but miss critical details in physical laboratory work.

They pipette frozen samples, contaminate DNA, fail to account for evaporating solvents, and vortex open containers.

Models doing Software and Mathematics have had the benefit of instant verification loops. Science is an intrinsically physical endeavor.

AI is going to radically change how we do science. We're building facilities where models can execute the entire research process and learn directly from reality.

This entire facility is run by AI.

Right now, everything you see, the equipment, the people, and the inventory is currently being controlled by GPT-6 Astra.

Let me show you how it works.

Hey Astra, synthesize Benzylammonium for me.

We'll let that progress for a bit.

Meanwhile, let me show you what Astra is working with.

Each bench in the facility has a laptop, instruments, and multiple camera feeds.

We've reverse engineered the firmware of each instrument so the model can talk through it over software.

Everything in the facility is perfectly represented in data, where it is, and how much is available down to the microliter.

The facility is a physical harness for R&D.

Models can control and inspect equipment, people, and inventory in order to achieve their goal.

Let's go check on the task we started earlier.

Dante got assigned to this task, and the model just prompted him to bring this reagent from the inventory.

We can see Dante find the right amine, scan it with the machine, and confirm that it made it to the bench.

Then, Dante's instructed to pipette liquid from the bottle into the vial, and then load the vial into the tray of the Opentrons.

The model operates the Opentrons and instructs Dante to verify the molecule via LC-MS.

And now we have a sample containing the molecule that we asked for.

So how good is AI at actually doing science?

Today, we're releasing SciUniverse, a growing collection of scientific benchmarks across biology, material science, and chemistry.

We built our first facility to understand what the models can do, the risks of them having a physical space to research, and where they have to improve.

And our ambition is to help AI make discoveries that are entirely new.