On June 30, 2026, Anthropic unveiled two products at once: Claude Science, an AI workspace built specifically for scientific research, and Claude Sonnet 5, its latest frontier model. Claude Science brings more than 60 scientific databases into a single environment and runs on the researcher's own computer or server. Sonnet 5 is designed to deliver performance close to the higher-end Opus 4.8 at a substantially lower price. Together, the launches mark a shift in which AI moves beyond chatbots toward drug discovery and more autonomous research support.

"Claude Science" Unifies 60-Plus Databases into One Research Environment

Claude Science is offered as an AI research workbench, a single workspace where researchers can carry out analysis, coding, and computing side by side. Until now, scientists have had to move between different databases and tools for each field, such as genomics, proteins, and chemistry, each with its own query language and file formats. Claude Science consolidates these fragmented tools into one place so that every stage of research can be handled without switching environments.

More than 60 scientific databases come pre-configured for areas including genomics, single-cell analysis, proteomics, structural biology, and cheminformatics. Unlike a typical chat AI, it can also render specialized artifacts on screen, such as three-dimensional protein structures, genome browser tracks, and chemical drawings.

A key design choice is that the app runs on the researcher's own environment rather than in a web browser. It can be installed wherever the data already lives, whether that is a laptop, a lab Linux machine, an HPC (high-performance computing) login node, or a cloud virtual machine. Raw datasets and computation stay local, and only the content included in prompts and responses is processed on Anthropic's side, making it easier to handle sensitive research data. Claude Science is not a new dedicated model; it runs on existing Claude models, including Opus 4.8. Development has been underway since October 2025.

Claude Science became available as a beta on June 30 and supports macOS and Linux. It is open to subscribers on the Pro, Max, Team, and Enterprise plans.

Drug Discovery Support and an "AI for Science" Program Start Alongside

On the same day, Anthropic said it is also pursuing its own pre-clinical drug discovery programs using AI. Eric Kauderer-Abrams, the company's head of life sciences, explained at a press conference that the focus will be on so-called neglected diseases and other areas that the traditional pharmaceutical and biotech landscape may not consider attractive targets. The aim is to direct AI toward fields that carry a real disease burden even though large players are reluctant to tackle them.

The company also prepared a support scheme for researchers. Anthropic will provide up to 30,000 USD (about 4.9 million yen) in credits each for up to 50 "AI for Science" projects. Applications are open through July 15, with award notifications expected by July 31. It functions as an entry point for trying Claude Science in real research settings.

"Claude Sonnet 5" Delivers Opus 4.8-Class Performance at a Lower Price

The other highlight is the new model, Claude Sonnet 5. Anthropic positions it as its "most agentic Sonnet yet." It can autonomously handle everything from planning to operating tools such as browsers and terminals, as well as sustained coding, at a level the company says rivals the higher-end Opus 4.8.

Pricing is notable. Sonnet 5 is designed to cost less to run than the previous Sonnet 4.6, and as an introductory rate it is offered at 2 USD (about 320 yen) per 1 million input tokens and 10 USD (about 1,600 yen) per 1 million output tokens. That rate runs through August 31, after which it moves to standard pricing of 3 USD (about 480 yen) per 1 million input tokens and 15 USD (about 2,400 yen) per 1 million output tokens. The result is a high-performance model available in a price range that developers can work with comfortably.

※1 USD = 162 JPY (as of July 1, 2026)

Safety was also addressed. Sonnet 5 is better at refusing malicious requests and resisting hijacking through prompt injection (the insertion of unauthorized instructions), and it reportedly shows lower rates of hallucination (output not grounded in fact) and excessive sycophancy than its predecessor, Sonnet 4.6. In addition, it launched with cyber safeguards that detect and block dangerous cyber use in real time, enabled by default. The approach reflects a stance of advancing performance and safety measures together, on the premise that higher general intelligence also raises risk.

Summary

Anthropic rolled out Claude Science, which consolidates specialized scientific research into a single environment, alongside Claude Sonnet 5, which offers Opus 4.8-class performance at a lower price. With a design that lets research data stay local, plus support spanning drug discovery and AI for Science, this announcement signals that AI is moving into a stage of delivering results in real research and industry. For researchers and developers alike, the range of options has widened considerably.