Anthropic release Claude ‘Opus 4.6’

On February 5, 2026, Anthropic officially released Claude Opus 4.6, its most advanced model to date. The launch was part of a coordinated “day of AI” that also saw OpenAI release GPT-5.3-Codex, leading to a frenetic series of benchmark-shattering updates within minutes of each other.

Opus 4.6 is positioned as a “hybrid reasoning” model, designed to bridge the gap between instant responses and deep, long-horizon cognitive tasks. It is specifically optimized for professional software engineering, complex financial analysis, and autonomous agent coordination.

1. Key Features & Architectural Breakthroughs

The 4.6 update introduces several “step-change” capabilities that distinguish it from its predecessor, Opus 4.5.

• 1 Million Token Context Window: For the first time, Opus supports a 1M token window (in beta), allowing it to ingest and maintain consistency across massive multi-million-line codebases or decades of financial filings in a single pass.
• Adaptive Thinking: The model now utilizes a dynamic reasoning engine. Instead of a fixed “thinking budget,” it automatically decides how much effort to apply based on the task complexity—spending minutes on a complex bug but responding instantly to a simple request.
• Agent Teams (Agent Swarm): This new native capability allows Claude to spin up and coordinate multiple “subagents” in parallel. Instead of one AI working sequentially, a “team” of agents can tackle different parts of a project (e.g., frontend, backend, and documentation) simultaneously.
• 128K Output Tokens: The output limit has been doubled to 128,000 tokens, enabling it to write entire applications or massive research papers without the need for multiple prompts.

2. Performance Benchmarks: The “SaaSpocalypse” Factor

The release of Opus 4.6 followed a period of intense market volatility (dubbed the “SaaSpocalypse”) where software and service stocks dropped as investors feared Anthropic’s new tools would replace existing enterprise software.

Note: While Opus 4.6 briefly held the record on Terminal-Bench 2.0, GPT-5.3-Codex eclipsed it with a score of 77.3% just 27 minutes later.

3. Enterprise & Product Integrations

Anthropic has moved away from a standalone chatbot model toward a deeply integrated enterprise ecosystem.

• Claude in PowerPoint & Excel: Now in research preview, Claude can natively edit pivot tables, modify slide masters, and build presentations that match a company’s specific brand guidelines and fonts.
• GitHub Copilot GA: Opus 4.6 reached general availability for GitHub Copilot on day one, allowing developers to use it as the primary engine for “vibe coding” sessions.
• Compaction API: A new beta feature that provides server-side context summarization, effectively enabling “infinite conversations” by summarizing old parts of a thread as you hit token limits.
• Data Residency: Users can now specify US-only inference (at a 1.1x price premium) to meet strict regulatory and compliance standards.

4. Pricing and Availability

• API Pricing: Remains at $5 per million input tokens and $25 per million output tokens for standard usage.
• Premium Context: For prompts exceeding 200,000 tokens, a higher tier of $10/$37.50 applies.
• Consumer Access: Available for Pro, Max, Team, and Enterprise users on Claude.ai.

Conclusion: The Era of “Vibe Working”

Anthropic CEO Dario Amodei and Head of Product Scott White have described Opus 4.6 as the beginning of the “Vibe Working” era—where the technical barrier to complex work is so low that non-experts can manage autonomous AI “squads” to build production-grade tools. With its massive context and adaptive reasoning, Opus 4.6 is less a “tool” and more a “digital colleague” capable of handling the full lifecycle of a project.