We review announcements from the official Anthropic news page and compile the most important news for Claude Code users. For the most up-to-date information, always refer to the official news.
February 2026
Vercept Acquisition — Strengthening Computer Use Capabilities (February 25)
Anthropic has acquired Vercept to enhance Claude's Computer Use capabilities — the ability for AI to see the actual screen and operate the mouse/keyboard.
Key Details: Vercept was a team researching technology for AI to see and interact with software like a human. They have joined Anthropic to focus on developing Computer Use features
Results: Sonnet 4.6's Computer Use score reached 72.5% on the OSWorld benchmark (a significant improvement from under 15% in late 2024)
Significance: Claude can now directly interact with actual application interfaces to perform complex tasks, beyond just writing code
ℹ️What is Computer Use?
Computer Use is a feature that allows AI to see the screen, click the mouse, and type on the keyboard, just like a human. For example, if you say "Calculate the sum of column A in this spreadsheet," the AI will open Excel and enter the formula itself. This enables handling GUI (graphical user interface) tasks that cannot be done through code alone.
Anthropic's Responsible Scaling Policy version 3.0 has been released. It embodies Anthropic's core principle of applying stricter safety standards as AI models become more powerful.
Claude Code Security Launch (February 20)
Claude Code Security has been released as a limited research preview. It is a new feature that scans codebases for security vulnerabilities and suggests fix patches.
How it works: Unlike traditional static analysis tools that only catch known patterns, Claude reads and reasons through code like a human security researcher to find complex vulnerabilities
Verification process: All findings go through multi-stage verification, where Claude attempts to prove or disprove its own findings to filter out false positives
Target audience: Prioritized for Enterprise and Team customers, with early access for open-source maintainers
Results: Using Opus 4.6, over 500 vulnerabilities were discovered in production open-source codebases — bugs that had gone undetected through decades of expert review
💡Static Analysis vs AI Analysis
Static Analysis: A method of inspecting code according to predefined rules. It excels at catching known patterns like "a password is exposed in the code," but may miss new types of vulnerabilities.
AI Analysis: By understanding the full context of the code — how data flows and how components interact — it reasons through and discovers complex vulnerabilities that rule-based approaches alone would miss.
Claude Sonnet 4.6 Release (February 17)
Claude Sonnet 4.6 has been released, delivering frontier-level performance across coding, agentic tasks, and professional workflows.
Model ID: claude-sonnet-4-6
Computer Use: Achieved 72.5% on the OSWorld benchmark, approaching human-level performance
Highlights: High performance relative to cost, fast response times
Claude Opus 4.6 Release (February 5)
Anthropic's most powerful model, Claude Opus 4.6, has been released.
Model ID: claude-opus-4-6
Pricing: $5 input / $25 output (per million tokens)
Context Window: First-ever 1 million token context window for Opus (beta)
Key Performance Improvements
Agentic Coding: Achieved the top score on Terminal-Bench 2.0. Plans more carefully, sustains longer agent tasks, and operates more reliably on large codebases
Reasoning Ability: Ranked #1 among all frontier models on Humanity's Last Exam (a complex interdisciplinary reasoning test)
Practical Capability: On GDPval-AA (an assessment of economically valuable knowledge work), scored approximately 144 Elo points ahead of GPT-5.2 and 190 points above the previous Opus 4.5
Search Ability: Achieved the best performance among all models on BrowseComp (a challenging information retrieval benchmark)
ℹ️Opus 4.6 Key Features at a Glance
Automatically focuses on difficult parts and handles easy parts quickly
Improved judgment on ambiguous problems
Maintains productivity even in long sessions
Enhanced code review/debugging ability to catch its own mistakes
If the model slows down from thinking too deeply, you can lower the effort to medium using the /effort command.
New Features Related to Claude Code
Agent Teams: Multiple agents can work as a team in Claude Code — Agent Teams Guide
Compaction: Via the API, Claude can summarize its own context to continue working longer without hitting token limits
Adaptive Thinking: Automatically adjusts the depth of Extended Thinking based on context
Effort Controls: Developers can fine-tune the balance between intelligence, speed, and cost
"Claude is a Space for Thinking" Declaration (February 4)
Anthropic officially announced its principle of never placing ads in Claude.
Core Message: Claude will remain a pure thinking and working space, free of advertising
Reason: AI conversations with users often cover deep and personal topics, and introducing ads could undermine trust
Principle: Ensuring that Claude's responses are never influenced by advertisers. "An environment where you never have to wonder whether the AI is genuinely helping you or subtly trying to sell something"
Other February 2026 News
Date
News
Details
Feb 23
Distillation Attack Prevention
Released technology to detect and prevent "distillation attacks" where other AIs use Claude's outputs as unauthorized training data
Feb 17
Anthropic-Rwanda MOU
Signed a Memorandum of Understanding (MOU) with the Rwandan government to apply AI in healthcare and education
Feb 17
Infosys Partnership
Joint development of AI agents for regulated industries such as telecommunications
Feb 16
Bengaluru, India Office Opening
Full-scale entry into the Indian market
Feb 13
CodePath Partnership
Introduced Claude to the largest university computer science program in the United States
How to Check for Updates
If you want to check for changes in the official news directly, use the methods below:
Official News Page
You can check the latest announcements on the Anthropic News page.
Using the Update Script
Running the update-checking script included in this project will automatically detect differences from the official documentation:
# Run the official documentation change detection scriptnode scripts/check-updates.mjs
💡What the Script Does
Fetches official documentation pages and extracts key terms and feature names
Compares them with previously checked content to find changes
If there are changes, it provides a summary of what has been updated
Found something missing from this guide?
If you discover a feature that exists in the official documentation but is not covered in this guide, please let us know through a GitHub issue and we will add it promptly.