- → On June 30, 2026, Anthropic launched Claude Sonnet 5presented as its most agentic model at this price.
- → Aggressive call rate: $2 in and $10 out per million tokens (until August 31).
- → In parallel, Claude Science : a multi-agent workshop for genomics, proteomics and chemistry, in beta.
- → A bet on the drug discoverybacked by a $200 million partnership with the Gates Foundation.
Two announcements on the same day, two directions. On June 30, 2026, Anthropic first launched Claude Sonnet 5, a new version of its mid-range model, designed to run agents at a lower cost. Then, more surprisingly, Claude Science: a tool designed not for the general public, but for laboratories. Deciphering a day that tells where Anthropic wants to go.
June 30 · double announcement
$2 / $10 · the Sonnet 5 rate
60+ · scientific connectors
$200 million · Gates Foundation partnership
01, Sonnet 5
An agent model, cheaper
Claude Sonnet 5 replaces Sonnet 4.6 in the Anthropic range. Its promise: to offer agent capabilities, plan a task, use tools, advance alone on several stages, which until now required a larger and more expensive model like Opus. On agentic code tests, Anthropic announces 63.2% for Sonnet 5, compared to 58.1% for Sonnet 4.6, getting closer to Opus 4.8 (69.2%).
But the biggest argument is the price. As an introductory price until August 31, Sonnet 5 costs $2 entry and $10 exit per million tokens, before increasing to $3 / $15. At this level, running agents continuously becomes significantly more affordable, which matters when an agent consumes millions of tokens per day.
Introducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.
02, Claude Science
A workshop for scientists
The other announcement is more unexpected. Claude Science is not a model, but a multi-agent workbench designed for research. The idea: instead of a single assistant, a coordinating agent who distributes the work to specialized agents, with a verification agent who rereads the results, tracks down erroneous citations and calculation errors.
The coordinating agent pilots the project and draws on more than 60 connectors and scientific bases (PubMed, ChEMBL, Benchling, Owkin’s pathology explorer).
Specialist agents can be tailor-made by the researcher for a specific task: genomics, proteomics, structural biology, chemistry.
The verifying agent inspects results, flags questionable citations, numerical errors, and discrepancies between code and figures. It’s the safeguard.
Available in beta for Pro, Max, Team, and Enterprise plans, Claude Science focuses on workflow rather than a new model: it builds on Anthropic’s existing models, but revamps the way they work together.
03, The health bet
Compress drug discovery
Why science? Because Anthropic sees this as an area where AI can really speed things up. The company announced a four-year, $200 million partnership with the Gates Foundation, focused on neglected diseases: polio, human papillomavirus (HPV), pre-eclampsia.
The objective is to shorten the first steps in the discovery of treatments, by computer screening of vaccine or therapeutic candidates before preclinical trials, which are long and expensive. Anthropic cites an internal case where AI helped move from nine protein targets to nine strong candidates for drug design. Be careful, however: these promises remain to be demonstrated on a large scale, and Anthropic does not communicate a verifiable quantified timetable.
04, What it says about Anthropic
Two audiences, one strategy
These two announcements share the same ambition. On the one hand, Sonnet 5 democratizes agents for developers and businesses, by cutting prices. On the other hand, Claude Science targets a niche but high-impact audience: researchers, pharma, biology.
Anthropic isn’t just looking to sell one more chatbot. The company positions itself as a production tool, for code as well as for science. It is also a response to competition: while OpenAI pushes its general public agents, Anthropic focuses on seriousness, verification and high-stakes areas.
To go further: Claude Fable 5 is back · Claude Opus 4.8 · The Claude model sheet · Compare AI models
Frequently asked questions
- What is Claude Sonnet 5?
- This is the new version of Anthropic’s mid-range model, launched on June 30, 2026 to replace Sonnet 4.6. It is designed to run agents (planning, use of tools, multi-step tasks) at a lower cost than Opus, with a call rate of $2 in and $10 in output per million tokens until August 31.
- What is Claude Science?
- A multi-agent workshop designed for scientific research, in beta. A coordinating agent distributes the work to specialized agents (genomics, proteomics, chemistry) and a verifying agent reads the results. It relies on more than 60 connectors and scientific bases such as PubMed, ChEMBL or Benchling.
- Is Claude Science a new model?
- No. Claude Science is a tool that reorganizes how Anthropic’s existing models work together, not a new model. It focuses on workflow and verification, with a dedicated agent tracking down misquotes and calculation errors.
- What is the connection with the Gates Foundation?
- Anthropic announced a four-year, $200 million partnership with the Gates Foundation, focused on neglected diseases (polio, HPV, pre-eclampsia), to accelerate the early stages of treatment discovery using AI.
- Is Sonnet 5 better than Opus 4.8?
- No, Opus remains more powerful: Anthropic announces 63.2% for Sonnet 5 on the agentic code, compared to 69.2% for Opus 4.8. The advantage of Sonnet 5 is its capacity-price ratio: it approaches Opus on certain agent tasks for much less.
- Can AI really accelerate drug discovery?
- This is the ambition, but it remains to be demonstrated on a large scale. AI can computer-screen candidates before preclinical trials, saving time upfront. Anthropic cites promising internal cases, without providing a verifiable quantified timetable.

