Anthropic and OpenAI opened a new front in their rivalry on Tuesday, both aiming at scientific research. Anthropic launched Claude Science, an AI workbench for researchers, while OpenAI released GeneBench-Pro, a benchmark for computational biology.

The same-day releases push the AI race beyond chatbots and coding into laboratory work. One company shipped a tool for scientists to use today. The other built a yardstick for how far the technology still has to go.

What Anthropic’s Claude Science Does

Claude Science brings the databases, code, and computing power scientists use into a single app. It connects more than 60 scientific databases across genomics, proteomics, and cheminformatics.

Claude Science is an app, not a new model. It lands while Anthropic’s most powerful Fable 5 and Mythos 5 models stay restricted under US export rules. Every result is auditable and traced back to the code that produced it.

The workbench extends a life sciences push Anthropic began in October 2025. In beta, the Allen Institute’s Jérôme Lecoq used it to compress reviews that once took up to two years.

Anthropic will also fund up to 50 research projects, with up to $30,000 in credits each.

OpenAI’s GeneBench-Pro Raises the Bar

Shortly after Anthropic’s Claude Science release, OpenAI released GeneBench-Pro. It tests whether AI agents can make the judgment calls that real biology research demands.

The benchmark contains 129 problems across genomics, quantitative biology, and translational medicine.

OpenAI’s strongest model, GPT-5.6 Sol, solved 28.7% of the problems at its highest reasoning level. That figure rose to 31.5% in Pro mode. The company’s earlier staggered GPT-5.6 release came at Washington’s request.

GPT-5 scored below 5% on the original GeneBench, while Anthropic’s Opus 4.8 reached 16% on the harder test.

Follow us on X to get the latest news as it happens

Two Strategies, One Race

The split reveals two paths to the same goal. Anthropic is shipping a product for daily lab use. OpenAI is measuring how reliably models reason through messy data.

Both launches also arrive as Chinese models gain ground in AI research. OpenAI’s own numbers temper the hype because its best model still fails most GeneBench-Pro tasks.

The pressure is both geopolitical and scientific. US export limits have already pushed Anthropic to weigh new host countries for its models.

Reviewers estimated each GeneBench-Pro problem would take a human expert 20 to 40 hours, costing thousands of dollars. OpenAI said its model finishes the same analysis for a few dollars.

Aubrey de Grey, a biomedical gerontologist, sees AI clearing key research bottlenecks even if broader gains take longer.

“What we’re going to see very very soon is that AI will make certain parts of the process, especially the development of drugs no longer rate limiting,” Aubrey de Grey, President and Chief Science Officer of the Longevity Escape Velocity Foundation, speaking on a BeInCrypto podcast.

De Grey cautioned that turning faster research into approved treatments still depends on regulation and public tolerance for risk.

Researchers Expect Faster Adoption

Some specialists argue the shift is already underway. Dr. Derya Unutmaz, a Professor of Immunology, told the same BeInCrypto panel that AI now outperforms his own judgment.

“I personally trust AI more than my own ideas in my field of 35 years.”

He expects that reliance to spread quickly across clinical practice.

“It is unethical and I believe that very soon it’s going to be malpractice not to use AI in medicine.”

That optimism still runs ahead of the benchmarks. The coming months will show whether scientists adopt these tools and whether GeneBench-Pro scores start to climb.

Source: BeInCrypto