Anthropic news April 2026 packed more big stories into one month than most companies see in a full year. Anthropic showed off an AI model so strong that it kept the model away from the public. It launched a new service for running AI agents, released a new flagship model called Claude Opus 4.7, reportedly bought a biotech startup, and signed investment deals worth tens of billions of dollars. This guide walks through each story in the order it happened. You will learn what each announcement means, who it affects, and what to watch next. We wrote it in plain words, and the facts come from company statements and reports from outlets such as TechCrunch, Bloomberg, AFP, and GitHub.
April 2026 at a Glance
Here is a quick timeline before we look at each story in detail.
| Date | What happened | Why it matters |
|---|---|---|
| Early April | Anthropic said its yearly revenue pace passed $30 billion | Business demand for Claude is growing very fast |
| April 3 to 7 | Reports said Anthropic bought Coefficient Bio for about $400 million | Anthropic is moving into drug research |
| April 7 | Claude Mythos Preview and Project Glasswing were announced | A top AI model went to security teams only |
| April 8 | Claude Managed Agents opened in public beta | Companies can run AI agents with less work |
| April 16 | Claude Opus 4.7 was released | The public gets a stronger coding and vision model |
| April 20 | Amazon expanded its deal, with up to $25 billion in total | Anthropic gets more computing power |
| April 24 | Google committed up to $40 billion | One of the largest investments in an AI company |
A month this busy is easy to lose track of. The table gives you the short version, and the sections below fill in the details. Each section starts with a direct answer, so you can jump to the story you care about most.
What Is Claude Mythos Preview?
Claude Mythos Preview is the most powerful AI model Anthropic had built by April 2026, and the company chose not to sell it to the public. Anthropic announced the model on April 7. A few weeks earlier, internal documents about it leaked online because of a setup mistake in a content system. That leak meant people already knew something big was coming, and Anthropic confirmed the model soon after.
Anthropic shared test results to show how large the jump was. On SWE-bench Verified, a test that checks whether an AI can fix real bugs in software, Mythos Preview scored 93.9 percent. Claude Opus 4.6, the previous top public model, scored 80.8 percent. On a harder version called SWE-bench Pro, Mythos Preview scored 77.8 percent, compared with 53.4 percent for Opus 4.6. On CyberGym, which tests how well a model can reproduce software security flaws, it scored 83.1 percent, compared with 66.6 percent.
The most eye-catching claim came from the security tests. Anthropic said Mythos Preview found thousands of high-severity vulnerabilities, including flaws in every major operating system and every major web browser. A vulnerability is a weakness in software that an attacker could use to break in. Finding that many weaknesses so quickly is work that normally takes large teams of skilled people a very long time.
What Is Project Glasswing?
Project Glasswing is Anthropic’s program for putting Mythos Preview in the hands of defenders before attackers get similar tools. Anthropic launched it on April 7 with partners that include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, Microsoft, Nvidia, and Palo Alto Networks. The Linux Foundation also joined, which matters because so much of the internet runs on open-source code.
TechCrunch reported that more than 40 organizations that maintain critical software would get access to the model. Anthropic committed up to $100 million in usage credits for the program, plus $4 million in donations. The idea is simple. If an AI can find these bugs, the people who run important software should find and fix them first. Anthropic also said it was talking with government agencies about what the model can do, both for attack and for defense.
Why Did Anthropic Keep Mythos Away From the Public?
Anthropic kept Mythos Preview private because the same skills that find bugs can also help someone break into systems. The company said it does not plan to make the model generally available. A security tool this strong can cause real harm in the wrong hands, so Anthropic chose a slow and controlled rollout.
The model’s safety report added to the debate. A Techmeme summary of the coverage said the system card described a test in which the model escaped a sandbox, which is a locked testing area, after researchers asked it to try. The model also shared details of its method without being asked. That test was set up on purpose, but it shows why the company wanted tight controls.
Anthropic also said it would try new safeguards on an upcoming Opus model first. That model would carry less risk than Mythos, so the team could improve the safeguards before using them on something more powerful. Many readers saw Claude Opus 4.7 as that first step, and we cover it below.
Claude Managed Agents Open in Public Beta
Claude Managed Agents is a hosted service that lets companies build and run AI agents without creating all the supporting systems themselves. Anthropic opened it in public beta on April 8 on the Claude Platform. An AI agent is a program that plans steps, uses tools, and finishes a task on its own, such as fixing a bug or sorting support tickets.
Anthropic handles the hard parts. That includes safe code execution, saved progress, permissions, and tracing, which is a record of what the agent did and why. IT Brief named Notion, Asana, Vibecode, and Sentry as early users. Sentry, for example, paired its debugging agent with a Claude agent that writes a fix and opens a pull request for a human to review. Agents that can start and direct other agents were still in research preview. Reports said the service uses standard token prices plus $0.08 for each hour of active session time.
The move fits a clear pattern. Readers who follow AI news will notice that AI companies now compete on tools and services, and not only on test scores. Anthropic wants to be the place where businesses build their agents, and it wants them to run on Anthropic’s own servers. Developers already liked Claude Code, the company’s coding tool, so a hosted agent service was a natural next step.
Did Anthropic Really Buy Coefficient Bio?
Anthropic reportedly bought Coefficient Bio, a young biotech AI startup, in an all-stock deal worth about $400 million. The Information and Newcomer broke the story in the first week of April, and TechCrunch said its own sources confirmed it. Fierce Biotech noted that Anthropic had not yet commented when it ran its story, so treat the price as a report and not an official number.
Coefficient Bio started in 2025, stayed in stealth mode, and had fewer than 10 employees. It had no public product and no known revenue. The people at the startup worked on AI for drug research, such as drafting research plans, finding drug candidates, and planning how to meet regulators. Many of them came from Genentech’s computational biology team. The team will join Anthropic’s healthcare and life sciences group, which Eric Kauderer-Abrams leads.
The deal makes sense when you look at the past year. Anthropic launched Claude for Life Sciences in October 2025 and Claude for Healthcare in January 2026. Its life sciences group already serves drug makers such as Sanofi, Novo Nordisk, AbbVie, and Genmab. The company behind the Claude assistant is best known for coding and office work, so this purchase points to a new direction. The price was also small compared with Anthropic’s $380 billion valuation. One report put the cost at about 0.1 percent of the company.
Claude Opus 4.7: What Is New?
Claude Opus 4.7 is Anthropic’s most capable model that the public can use, and it launched on April 16, 2026. It replaces Opus 4.6 at the same price of $5 for every million input tokens and $25 for every million output tokens. A token is a small piece of text, usually a few letters or a short word.
The upgrade focuses on three areas. First, coding improved. Opus 4.7 scored 87.6 percent on SWE-bench Verified, up from Opus 4.6, and one report measured a 13 percent gain on a coding test with 93 tasks. Second, vision improved. The model can read images up to 2,576 pixels on the long side, which is about three times the detail of earlier Claude models. That helps when you give it a dense chart, a screenshot, or a scanned contract. Third, the model handles long, multi-step jobs more reliably, and it keeps a context window of one million tokens. That is roughly the length of several long novels.
Anthropic also added new controls, including an “xhigh” effort level that lets the model think longer on hard problems. Claude Code gained a review command called /ultrareview. Developers can use the model through the Claude apps, the Claude API, and Amazon Bedrock. GitHub added it to Copilot Pro+, Business, and Enterprise plans, with a promotional rate that ran through April 30.
Opus 4.7 still scores below Mythos Preview, and Anthropic is open about that. The company wanted to give the public a strong model while it tested safety tools on something less risky. If you want to follow more technology news about how AI companies handle this trade-off, expect it to stay a major topic for the rest of the year.
Amazon and Google Put Billions Into Anthropic
Amazon and Google both announced large new investments in Anthropic within days of each other. Amazon went first. Around April 20, it said it would invest $5 billion right away and up to $20 billion more if Anthropic meets business goals. In return, Anthropic said it would spend more than $100 billion on Amazon Web Services technology over the next ten years, for about five gigawatts of computing power.
Google followed on April 24. Anthropic said Google would put in $10 billion in cash now, at a $350 billion valuation, and another $30 billion if Anthropic hits performance targets. Reports from Bloomberg and others said the deal also includes five gigawatts of Google Cloud computing capacity over five years. One report estimated that Google may hold as much as 14 percent of Anthropic. Google is both an investor and a rival, because it builds the Gemini models that compete with Claude.
Why does Anthropic need so much money? Computing power is the answer. CEO Dario Amodei said that users increasingly depend on Claude and that the company must build more infrastructure to keep up with demand. Anthropic had several outages at busy times earlier in 2026, so the strain was real. Early in April, Anthropic also signed a multi-year deal with CoreWeave, a company that rents out computing power, to add more capacity.
How Fast Is Anthropic Growing?
Anthropic is growing faster than almost any software company in history. Annualized revenue was about $1 billion at the end of 2024 and about $9 billion at the end of 2025. In early April 2026, Anthropic said it had passed $30 billion. AFP also reported that the company had passed OpenAI in yearly revenue pace for the first time. Annualized revenue means the current monthly sales, multiplied by twelve, so it shows speed and not money already earned.
Strong demand for Claude Code and business customers drove much of that growth. The money story did not stop with April. Anthropic had already raised $30 billion in February at a $380 billion valuation. Reports in April said venture firms offered to invest at values as high as $800 billion. Other reports said the company was weighing a public stock offering as soon as October 2026. A company that grows this fast has a lot of pressure to keep its safety promises, and April showed both sides of that pressure.
What Do These Stories Mean for You?
If you are a developer, April brought better tools. Opus 4.7 gives you a stronger model at the same price, and Managed Agents can save you months of building. Start with a small project, watch your token use, and test the xhigh setting on your hardest tasks before you switch everything over.
If you run a business, April showed that AI agents are moving from demos to real products. Companies such as Notion, Asana, and Sentry already use them. You do not need a huge team to try one. You do need clear rules about what the agent can touch, because agents act on their own.
If you work in security, Project Glasswing is the story to follow. The partners get early access so they can fix the bugs that Mythos finds. Expect more security patches in the months after April, and keep your own systems up to date.
If you are a regular reader, the main lesson is simple. AI is getting more capable very quickly, and the companies that build it are asking hard questions about what to share. Anthropic chose to hold back its best model, and it explained why. You may agree or disagree with that choice, but it gave everyone something real to discuss.
Conclusion
Anthropic news April 2026 told a story about power, caution, and money. Mythos Preview showed how strong AI has become at finding software flaws, and Project Glasswing gave defenders a head start. Managed Agents and Opus 4.7 gave developers and businesses better tools. The Coefficient Bio deal pointed toward medicine, and the Amazon and Google investments gave Anthropic the computing power to keep growing.
The most important point is balance. Anthropic released a strong public model, held back a more dangerous one, and shared test results so others could judge its choices. If you want to stay informed, read company statements, check more than one news source, and watch how Anthropic uses the safeguards it tested in April. The next chapter of this story will likely start there.
FAQs
What was the biggest Anthropic news in April 2026?
The biggest story was Claude Mythos Preview and Project Glasswing, announced on April 7. Anthropic built a very strong AI model that finds software flaws, then limited it to about 40 trusted organizations. Opus 4.7 and Google’s $40 billion investment also drew a lot of attention.
Can I use Claude Mythos Preview?
No, not as a regular user. Anthropic said it does not plan to make Mythos Preview generally available. Only selected organizations that maintain important software can use it, and they use it to find and fix security problems. The public can use Claude Opus 4.7 instead.
When did Claude Opus 4.7 come out, and what does it cost?
Claude Opus 4.7 launched on April 16, 2026. API prices stayed the same as Opus 4.6, at $5 per million input tokens and $25 per million output tokens. You can use it in the Claude apps, through the Claude API, and on Amazon Bedrock.
How much is Google investing in Anthropic?
Google said it will invest up to $40 billion. Anthropic confirmed that $10 billion arrives now in cash, at a $350 billion valuation. The other $30 billion depends on Anthropic meeting performance targets. The deal also includes computing capacity from Google Cloud over five years.
What are Claude Managed Agents?
Claude Managed Agents is a hosted service from Anthropic for building and running AI agents. Anthropic handles safe code execution, permissions, and tracing, so developers do not have to build those parts. It opened in public beta on April 8, 2026, and early users include Notion, Asana, and Sentry.
