Claude Updates and Changelog

Anthropic released Claude Fable 5, its first Mythos class model for general availability. Fable 5 uses the same architecture as the restricted Mythos model from April, but ships with safety classifiers that route high risk queries in cybersecurity and biology to Claude Opus 4.8 instead. The model scores 80.3% on SWE Bench Pro (Opus 4.8 scored 69.2%, GPT 5.5 scored 58.6%) and sets new highs in knowledge work, vision, and tool use, with its lead widening as tasks get longer and harder. Stripe reported that Fable 5 completed a codebase wide migration on 50 million lines of Ruby in one day, work that would have taken an engineering team two months. Pricing is $10 and $50 per million input and output tokens, double Opus 4.8. Included on Pro, Max, Team, and Enterprise plans at no extra cost through June 22.

Alongside Fable 5, Anthropic shipped Claude Mythos 5 as an upgrade to the restricted Mythos Preview for Project Glasswing partners. Mythos 5 is the same underlying model as Fable 5 but with safety classifiers removed in specific domains, making it the most capable cybersecurity model in the world. Beyond security, Anthropic reported that Mythos 5 is the company's first model to consistently produce novel scientific hypotheses that researchers preferred over Opus class outputs 80% of the time in blinded comparisons. In genomics, the model autonomously assembled single cell data across 138 animal species and trained a custom machine learning model that outperformed a recent Science journal publication while being 100 times smaller. Anthropic plans to expand Mythos 5 access through a broader trusted program beyond the current Glasswing cohort.

Anthropic released Claude Opus 4.8, its most capable generally available model, just 41 days after Opus 4.7. It improves on its predecessor with sharper judgment on ambiguous and high stakes decisions, more honesty about its own progress and limitations, and stronger long horizon autonomy. Agentic coding rose from 64.3% to 69.2% and multidisciplinary reasoning with tools climbed from 54.7% to 57.9%. A new Fast Mode runs the full model at roughly 2.5 times the speed, and the model is roughly four times less likely to let coding errors pass unremarked. Pricing is unchanged from Opus 4.7 at $5 and $25 per million input and output tokens, with Fast Mode at $10 and $50. Opus 4.8 uses high effort by default and adds extra and max effort levels for difficult, long running work.

Alongside Opus 4.8, Anthropic launched Dynamic Workflows in research preview, letting Claude run hundreds of parallel subagents to take on larger tasks in Claude Code, including codebase scale migrations across hundreds of thousands of lines of code from kickoff to merge. A new effort control lets users decide how much reasoning Claude applies to a response, with higher settings thinking longer and lower settings returning quicker, cheaper answers. Effort controls are available in Claude Code, the browser, and Claude Cowork.

Claude Security (previously Claude Code Security) entered public beta for Claude Enterprise customers. Powered by Opus 4.7, the tool scans entire codebases for vulnerabilities and generates targeted patches without requiring API integration or custom agents. New features include scheduled and targeted scans, directory level scanning, dismissible findings with documented reasons for audit trails, CSV and Markdown export, and webhook integrations with Slack and Jira. Technology partners CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, and Wiz are embedding Opus 4.7 into their security platforms.

Claude Design launched alongside Opus 4.7 as a new Anthropic Labs product. It allows users to collaborate with Claude to create visual outputs including designs, prototypes, slides, and one pagers. The tool reads a team's codebase and design files to build a reusable design system automatically. Results export to Canva, PDF, PPTX, or standalone HTML, and designs can be handed off to Claude Code for implementation. Available in research preview for Pro, Max, Team, and Enterprise subscribers.

Anthropic released Claude Opus 4.7, its most capable generally available model at the time. Key improvements include 87.6% on SWE bench Verified (up from 80.8%), a 3.3x increase in maximum image resolution to 2,576 pixels, a new xhigh effort level for Claude Code, task budgets for controlling agentic token spend, and self verification behavior where the model proactively tests its own outputs before reporting results. The model uses a new tokenizer that can produce up to 35% more tokens for the same input text, which affects effective pricing despite the sticker price remaining at $5 and $25 per million input and output tokens.

Claude Cowork gained computer use capabilities, allowing it to open files, run developer tools, point, click, and navigate the user's screen to perform tasks autonomously. Alongside this, Pro and Max plan users can now dispatch tasks to Cowork from their phone via a persistent agent thread on Claude for iOS and Android, enabling remote task management without being at a computer.

Anthropic announced Claude Mythos Preview, described as more powerful than Opus 4.7 across every benchmark category. Access is restricted to 11 partner organizations through Project Glasswing for defensive cybersecurity workflows only. The UK AI Security Institute reported Mythos solved a 32 step cyber range called "The Last Ones" 3 out of 10 times, the first model to complete it end to end. Anthropic has stated it does not plan to make Mythos Preview generally available, though it expects to bring Mythos class models to all customers in the coming weeks once safeguards are complete.

Claude gained the ability to create custom charts, diagrams, and other visualizations rendered directly inline in conversation responses. This complements the existing Artifacts feature by allowing Claude to generate SVG and HTML based data driven visuals on the fly without requiring users to open a separate preview panel, making analysis workflows more interactive.

Anthropic raised the max_tokens cap to 300,000 on the Message Batches API for Claude Opus 4.6 and Sonnet 4.6. The feature requires a beta header and targets long form content generation, structured data extraction, and large scale code generation tasks where single turn output length was previously a bottleneck. Standard synchronous API calls retain their existing output limits.

Microsoft announced that Claude Sonnet would be available to Microsoft 365 Copilot users, marking Claude's first integration into the Microsoft productivity suite. This gives enterprise users access to Claude's capabilities directly within Word, Excel, PowerPoint, and Outlook workflows alongside Microsoft's own models, with users able to select Claude as their preferred model within the Copilot interface.

Two major API capabilities entered public beta. Managed Agents provides a fully managed harness for running Claude as an autonomous agent with secure sandboxing, built in tools, and server sent event streaming. The Advisor Tool pairs a faster executor model with a higher intelligence advisor model for strategic guidance during generation, allowing long horizon agentic workloads to approach advisor quality at executor model cost.

Claude Sonnet 4.6 shipped with near Opus level performance at Sonnet pricing ($3 and $15 per million tokens). It achieved 79.6% on SWE bench Verified, closing the gap with Opus 4.6 to just 1.2 percentage points. The model includes 1M token context window support at standard pricing with no beta header required, improved computer use for browser navigation and form filling, and stronger instruction following.

Opus 4.6 launched with the first generally available 1M token context window for an Opus model, an 83% improvement on ARC AGI 2 reasoning (68.8% vs 37.6%), and native agent team capabilities that allow splitting larger tasks across multiple coordinated agents. METR estimated its 50% task completion time horizon at 14 hours and 30 minutes, the longest of any model at the time. Claude in PowerPoint also launched alongside this release.

Enterprise plans became available for self serve purchase with no sales conversation required. A single seat type includes access to Claude, Claude Code, and Cowork with unified billing. Alongside this, the Enterprise Analytics API launched, providing programmatic access to usage and engagement data aggregated per organization per day, giving IT teams visibility into adoption patterns.

Claude Cowork launched as a research preview for Max plan users on macOS. Built by a four person team in 10 days using Claude Code itself, Cowork brings agentic capabilities to the Claude desktop app for knowledge work beyond coding. It runs locally in an isolated VM with direct access to local files and MCP integrations, enabling multi step file and task management workflows.

Claude Opus 4.5 became the first model to break the 80% barrier on SWE bench Verified with a score of 80.9%. The release also introduced Infinite Chats, which automatically summarizes earlier messages when a conversation approaches context limits instead of returning length limit errors. Improvements focused on coding and workplace tasks like producing spreadsheets. Opus 4.5 triggered a viral wave of Claude Code adoption over the winter holidays.

Claude for Excel entered beta for Max, Team, and Enterprise users. The add in supports pivot tables, charts, file uploads, and a shortcut to open the full Claude app from within Excel. It allows Claude to read, analyze, and manipulate spreadsheet data directly inside the Microsoft Office workflow without switching between applications.

Anthropic released Claude Haiku 4.5, its fastest and most cost efficient small model. Haiku 4.5 matches Claude Sonnet 4's performance on coding, computer use, and agent tasks while operating at the Haiku price tier. It became the default model for Claude in Chrome and the recommended option for high volume, latency sensitive deployments.

The Claude Code VS Code extension launched in beta, bringing inline diffs, a dedicated sidebar panel, and graphical interaction to developers who prefer IDE workflows over the command line. The extension includes keyboard shortcuts, plan review capabilities that let developers approve changes before execution, and direct integration with the VS Code terminal.

Claude Sonnet 4.5 launched with 77.2% on SWE bench Verified and was described as the world's best model for real world agents, coding, and computer use. The release included 30+ hour focus windows for sustained autonomous work and achieved a 0% error rate on Replit's internal benchmark. Sonnet 4.5 also brought the 1M token context window to the Sonnet tier in beta.

Claude Opus 4.1 shipped as an incremental update to the original Claude 4 with enhanced performance on complex reasoning, analysis, and creative tasks. The release also introduced the ability for Opus models to end conversations that remain persistently harmful or abusive as a last resort after multiple refusals. Opus 4.1 served as a drop in replacement with no API changes required.

Claude in Chrome launched as a Google Chrome extension enabling Claude to directly control the browser. Users can have Claude navigate web pages, fill forms, click buttons, take screenshots, and perform multi step browser tasks. The feature expanded to Max plan subscribers with scheduled recurring tasks and multi tab support in subsequent updates.

Anthropic launched the Claude 4 generation with both Opus 4 and Sonnet 4. Opus 4 achieved 72.5% on SWE bench and 43.2% on Terminal bench, becoming the leading coding model at launch. Both models support hybrid reasoning with extended thinking and tool use, parallel tool calls, and improved instruction following. Opus 4 was classified as Level 3 on Anthropic's safety scale. Claude Code became generally available alongside this release with GitHub Actions integration and native VS Code and JetBrains support.

Four new API capabilities launched alongside Claude 4: the code execution tool for running sandboxed code, MCP connectors for integrating external tools using the Model Context Protocol, the Files API for uploading and referencing documents, and prompt caching for up to one hour. Together these capabilities formed the foundation for building production grade AI agents on the Claude API.

Anthropic introduced the Max plan with two tiers: Max 5x at $100 per month and Max 20x at $200 per month, providing 5x or 20x the usage limits of the standard Pro subscription. Max subscribers get priority access to new models and features, combining Claude desktop, mobile apps, and Claude Code in one subscription. The plan targets professionals hitting Pro plan usage limits during extended work sessions.

Claude gained web search capabilities, starting with paid users in the United States and expanding to free users in May 2025. The feature allows Claude to search for and cite current information from the web during conversations, addressing one of the most requested capabilities: access to information beyond the training data cutoff.

Claude 3.7 Sonnet introduced hybrid reasoning with extended thinking, allowing the model to pause and think step by step before responding. This significantly improved performance on complex math, science, and multi step coding problems. The model maintained instant response mode for simple queries while activating deeper reasoning automatically for harder tasks.

Claude Code launched as a research preview: an agentic command line tool for delegating coding tasks from the terminal using natural language. Initial capabilities included file editing, bash command execution, and git workflow management. Though limited at launch, it set the foundation for what would become Anthropic's fastest growing product surface, eventually driving a reported 5.5x revenue increase by July 2025.