On June 30, 2026, Anthropic announced "Claude Science," an AI workbench for scientists. Rather than a new model, the company is betting on integrating existing Claude into research workflows to win over the scientific field.
June 30, 2026 · Anthropic
Claude Science Bets on Workflow, Not a New Model
Anthropic launches an AI workbench that wires existing Claude models into the scientist's daily research pipeline — wagering that better integration, not raw model power, will win over the lab.
60+
scientific databases & tool integrations via MCP
$30K
credit program for postdocs & grad students, up to 50 projects
0
new dedicated model — runs on Opus 4.8 as-is
Protocol QA benchmark
Understanding of experimental protocols — the Sonnet 4.5-based version edges past the human-expert baseline.
Three paths into science
Anthropic
Existing model + workflow integration. Beta now, on-prem support.
OpenAI
Dedicated fine-tuned model (GPT-Rosalind), limited preview with safety review.
Google DeepMind
In-house foundational science models — AlphaFold, AlphaGenome, Gemini for Science.
How the workbench runs
Assistant as "project manager"
→
Spawns sub-assistants + field toolkits
→
Fact-checker AI verifies citations
→
Reproducible figures + source code attached
Why scientists welcome it
End-to-end support in one environment
Built-in reproducibility & plain-language figure editing
On-prem option eases data-privacy fears
A genome browser built in days; years of work compressed
The open questions
Fact-checker uses the same model — not an independent truth source
Hallucinations & fabricated citations persist
Expert-level tasks still need human oversight
"Integration quality, not raw model power, is the key"
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…