Ground truthfor biology AI.

We're building a library of expert-written biology tasks with locked, verified answers, packaged as training environments and private evaluations for the teams building AI scientists.

Every task will be re-solved by a second expert, tested against frontier agents, and cleared by a biosecurity review before release.

Cell surface illustration

Biology agents can only get as good as the tasks they learn from.

Agents can now run whole analyses: read the literature, pull the data, write the code, draw a conclusion. Teaching them to do it well takes tasks that are hard, realistic and checkable, and biology has very few.

  • Most biology tasks are quiz-sized. Real research takes many steps and many tools.
  • Answers are hard to verify. In biology, the right answer is often only provable by experiment.
  • Everyone grades their own homework. Most benchmarks are built by the companies whose agents they score.

What we're building

A library of biology tasks you can trust the answers to.

Each task will be a complete research problem: a question a scientist would actually ask, real data, a sandbox with the standard tools, and an answer that has been independently confirmed.

Training environments

Containerised, multi-step tasks with automatic graders, built for reinforcement learning. Agents will earn credit for sound method, not just a lucky final answer.

Private evaluations

Held-out suites that will never be published or reused for training, refreshed every quarter, so scores show real capability, not memorisation.

Public scorecard

A planned annual, neutral ranking of biology agents on a shared subset.

Verification

How every task will be verified

  1. An expert writes it.

    A practising PhD or postdoc builds the brief, the data and the reference solution.
  2. A second expert re-solves it, blind.

    Any disagreement is resolved before the task moves on.
  3. We lock the truth.

    Where analysis alone can't settle an answer, we commission the experiment.
  4. Frontier agents attack it.

    If they solve it too easily or can fool the grader, it goes back for rework.
  5. A biosecurity review clears it.

    Every task is screened for dual-use risk with an outside adviser.
  6. It ships with its answer locked.

    We'll never change an answer after release without a public erratum.

Why Cognita Bio

CriterionTypical sources of biology tasksCognita Bio
Task sizeSingle-answer questionsStudy-scale, multi-tool analyses
Ground truthOne author's answerBlind re-solve, experiment-backed where needed
GradingFinal answer onlyFinal answer plus checks on each step
NeutralityOften built by an agent vendorIndependent: we don't build or sell an agent
SafetyVariesEvery task cleared by a biosecurity review

The first release

Our first release will focus on two areas:

Single-cell analysis

Clustering, annotation, differential expression, trajectory analysis

Target identification

Genetic evidence, pathway reasoning, literature synthesis

We're building study-scale tasks across both areas, released in batches as each one passes verification. We won't rush a task to hit a number. Imaging and structural biology will follow.

Design partners

Help shape the first release.

We're working with a small number of design partners: model builders and biology agent platforms who want tasks built around their hardest problems.

Design partners get:

  • Input on which skills and workflows the first tasks cover
  • The first batch of tasks to run their models on, with a report on where those models fail
  • Early access to the private evaluation sets
  • Founding-partner pricing

Who it’s for

Model builders.

Post-training and evaluation teams training general models to do real science.

Biology agent platforms.

Teams who need an independent measure of their agent, and better tasks to train it on.

Our commitments

Neutral by design.

We don't build agents, so we'll never compete with the teams we serve.

Answers stay locked.

A held-out task will never be released, resold for training or reused.

Safety first.

We won't build tasks that teach harmful capabilities, whoever asks.

Errors in the open.

If we get an answer wrong, we'll fix it publicly and tell every customer who has the task.

FAQ

When will tasks be available?

The first batch goes to design partners. General access follows once the first release passes verification.

What formats will you deliver in?

We're planning Docker-based environments and a hosted evaluation API, and we'll confirm formats with design partners first.

Can we get tasks no one else has?

We plan to offer time-limited exclusive tasks. Tasks in the public scorecard set will never be exclusive.

Who writes the tasks?

A vetted network of PhD and postdoctoral scientists, with each task reviewed by a second expert in the same field. If you're a scientist interested in writing tasks, email us at hello@cognita.bio.

How do you handle biosecurity?

Every task will go through dual-use screening with an outside biosecurity adviser before release. Tasks that fail won't be built.

Do you sell to pharma?

Not at first. We're focused on the teams building AI for science.

Find out where your agent breaks.

Join as a design partner and be among the first to run your models on verified biology tasks.

Talk to the team: hello@cognita.bio

Apply to be a design partner

Tell us about your team and the problems you'd want tasks built around. Sending opens your email app with your application filled in, addressed to hello@cognita.bio.