Anthropic releases Claude Opus 4.7 with stronger coding and cyber safeguards
TL;DR
Anthropic launched Claude Opus 4.7 on April 16, positioning it as an improvement over Opus 4.6 (February's release) for agentic coding, instruction-following, and real-world work. The model ships with automatic safeguards that detect and block prohibited or high-risk cybersecurity requests.
What changed
What shipped
Claude Opus 4.7 is Anthropic's new generally-available flagship model. It succeeds Claude Opus 4.6 (released in February 2026) as the default for most Claude Pro, Max, Team, and Enterprise users.
Anthropic's own framing: 4.7 outperforms 4.6 across industry benchmarks for agentic coding, multidisciplinary reasoning, scaled tool use, and agentic computer use.
The Mythos context
4.7 arrives alongside a separate, more capable model called Claude Mythos Preview, which Anthropic is not making generally available. Mythos is deployed to a select group of companies as part of Project Glasswing, Anthropic's new cybersecurity initiative.
Anthropic describes Opus 4.7 as "less broadly capable" than Mythos, specifically in its cyber capabilities. The company experimented with efforts to "differentially reduce" Opus 4.7's cyber capabilities during training so that the general-availability model ships without the offensive capability ceiling Mythos has.
Built-in cyber safeguards
"We are releasing Opus 4.7 with safeguards that automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses," Anthropic said at launch. This is the first major Claude release where the cyber safeguards are called out as a primary product feature.
Deployment context
The launch follows a sequence of high-profile meetings between the Trump administration, tech CEOs, and bank chief executives about security risks of powerful AI models. Project Glasswing is Anthropic's response: a specialised, access-restricted deployment track for cybersecurity-sensitive work.
4.7 is the general-purpose model most users will interact with. Mythos is the cyber-specialised tier available to Project Glasswing participants only.
Availability
Claude Opus 4.7 rolled out to all Claude Pro, Max, Team, and Enterprise subscribers on April 16. Available through the Claude app, Claude.ai, the API, Claude Code, and Claude Design (Claude Design uses Opus 4.7 as its vision model).
Who this matters for
- Vibe Builder: Your default Claude model got meaningfully better at agentic coding and tool use. Expect fewer retries per complex task and better handling of multi-step workflows.
- Basic User: If you use Claude for writing, research, or general chat, you get better reasoning and instruction-following with no setting to change. Just keep using Claude normally.
- Developer: 4.7 is the new default on the API and in Claude Code. Agentic computer use and scaled tool use benchmarks improved over 4.6; worth revisiting any agent workflows that previously hit capability walls.
What to watch next
The most interesting detail is that 4.7 is Anthropic's general-availability model by design, not by default. They trained an upstream model, decided it was too cyber-capable for broad release, and trained a less-capable variant for public deployment. That is a significant departure from the usual "train a model, ship it widely" pattern. The decision structure matters: Anthropic is acting as the capability gatekeeper for frontier models in a way OpenAI and Google have not publicly committed to.
For most users, 4.7 is the practical benefit. It is the new default for Claude Pro, Max, Team, and Enterprise users. The upgrade to agentic coding and tool use is the operationally useful bit. If you use Claude for coding, expect fewer retries per task and better handling of multi-step workflows.
The Mythos tier is a lot more interesting to watch strategically. By creating a restricted-access variant exclusively for cybersecurity work, Anthropic is testing whether the AI industry can support tiered capability deployment, where the most powerful models go to specific, accountable use cases rather than the general public. If this works, expect other frontier labs to copy the pattern; if it breaks (leaks, misuse, or Mythos capabilities getting replicated open-source), expect a harder regulatory conversation.
The Project Glasswing context also explains why 4.7 is being positioned as "safe and capable" rather than "frontier and powerful." Anthropic is threading a needle: ship meaningful capability improvements for general users while not escalating the arms race on cyber capabilities.
by Harsh Desai
About claude
View the full claude page →All claude updatesMore from claude
- LaunchAnthropic launches Claude Design for prototypes, slides, and design systems
Anthropic shipped Claude Design, a new Anthropic Labs product that turns conversation into polished visual work: interactive prototypes, wireframes, pitch decks, and on-brand marketing collateral. Powered by Claude Opus 4.7 and available in research preview to Pro, Max, Team, and Enterprise subscribers.
- FundingAmazon Commits Another $5B to Anthropic, Locks In 5GW of Compute
Amazon announced on 20 April 2026 an additional $5 billion investment in Anthropic with an option for up to $20 billion more, securing 5 gigawatts of Trainium compute for Claude training and inference. Anthropic pledges over $100 billion of AWS spend across the next decade.