# Claude Sonnet 4.6 Just got released! Source: https://www.youtube.com/watch?v=RttuBBZUvv0 Recap page: https://rapidrecap.app/video/RttuBBZUvv0 Generated: 2026-02-17T20:34:57.629+00:00 --- ## Quick Overview Anthropic released Claude Sonnet 4.6 on February 17, 2026, which significantly improves performance across coding, computer use, long-context reasoning, agent planning, knowledge work, and design, achieving scores nearly equal to Opus 4.6 in several agentic benchmarks while maintaining the same pricing as Sonnet 4.5 ($3/$15 per million tokens) and introducing a 1M token context window in beta. **Key Points:** - Claude Sonnet 4.6 is positioned as the most capable Sonnet model yet, offering a full upgrade across skills including coding, computer use, reasoning, and design. - Sonnet 4.6 matches or nearly matches the performance of the previous frontier model, Opus 4.5, in agentic financial analysis (63.3% vs 60.1%) and closely rivals Opus 4.6 in several agentic tasks. - The new model features a 1M token context window in beta, and developers using Claude Code showed a strong preference (70% of the time) for Sonnet 4.6 over Sonnet 4.5. - Pricing remains the same as Sonnet 4.5, starting at $3/$15 per million tokens for input/output, which is cheaper than Opus 4.6. - On the Vending-Bench Arena, Sonnet 4.6 significantly outperforms Sonnet 4.5, achieving balances over $5,000 compared to under $2,000 after 300 days of simulation. - Sonnet 4.6 scores 72.5% on OSWorld-Verified computer use benchmarks, showing continuous improvement over previous Sonnet versions (Sonnet 4.5 scored 61.4%). - The free tier is upgraded to Sonnet 4.6 by default, now including file creation, connectors, skills, and context compaction. ![Screenshot at 00:07: The comparison table highlighting that Claude Sonnet 4.6 \(59.1% on Agentic terminal coding\) is now matching or outperforming previous models like Opus 4.6 \(65.4%\) and Sonnet 4.5 \(51.0%\) in several agentic benchmarks.](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-00-07.jpg) **Context:** The video announces the release of Claude Sonnet 4.6 by Anthropic on February 17, 2026, positioning it as a major upgrade to their mid-tier model. The presentation uses benchmark comparison tables (Agentic coding, tool use, reasoning) and simulation graphs (Vending-Bench Arena) to demonstrate Sonnet 4.6's performance gains over its predecessor, Sonnet 4.5, and its closer parity with the larger Opus model, while also highlighting new features like a larger context window and improved tool use capabilities. ## Detailed Analysis Claude Sonnet 4.6 was introduced as Anthropic's most capable Sonnet model, representing a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. A key feature is the 1M token context window, available in beta. Performance benchmarks show significant gains: Sonnet 4.6 scores 79.6% on Agentic coding (up from 77.2% for Sonnet 4.5) and 72.5% on Agentic computer use (up from 61.4%). In agentic tool use, Sonnet 4.6 achieves 91.7% (Retail) and 97.9% (Telecom), closely matching Opus 4.6 (91.9% and 99.3%). Crucially, in agentic financial analysis, Sonnet 4.6 scores 63.3%, better than Opus 4.5 (60.1%). In competitive leaderboards like ARC-AGI-2, Sonnet 4.6 (Score 54.2%, Cost $15.72) performs better than GPT-5.2 Pro (High) and is close to Opus 4.6. Pricing remains unchanged from Sonnet 4.5, starting at $3/$15 per million tokens, making it significantly cheaper than Opus 4.6. Furthermore, the free tier now automatically uses Sonnet 4.6 and includes file creation, connectors, skills, and context compaction. API enhancements include web search and fetch tools that can execute code to filter results, alongside code execution, memory, programmatic tool calling, tool search, and tool use examples. ### Performance Benchmarks - Sonnet 4.6 achieves 79.6% in Agentic Coding and 72.5% in Agentic Computer Use - Sonnet 4.6 scores 63.3% in Agentic Financial Analysis, surpassing Opus 4.5 (60.1%) - Sonnet 4.6 performance is closely aligned with Opus 4.6 in tool use benchmarks (e.g., 99.3% Telecom vs 99.3%) ### Key Features - Features a 1M token context window in beta - API tools now automatically filter and process search results via code execution - New capabilities include adaptive thinking, extended thinking, and context compaction in beta ### Pricing & Availability - Pricing remains the same as Sonnet 4.5 ($3/$15 per million tokens) - Available on Free and Pro plans, Claude Cowork, Claude Code, and API - Free tier upgraded to Sonnet 4.6 by default with added features like file creation and skills ### Agentic Behavior & User Preference - Users preferred Sonnet 4.6 over Sonnet 4.5 70% of the time in Claude Code testing - Users rated Sonnet 4.6 less prone to overengineering and laziness than Opus 4.5 - Sonnet 4.6 outperforms Sonnet 4.5 in Vending-Bench Arena simulation over 300 days ### Competitive Landscape (ARC-AGI-2) - Sonnet 4.6 (Score 54.2%, Cost $15.72) is positioned below Opus 4.6 (Score 72.9%, Cost $38.99) but performs better than GPT-5.2 Pro (High) ### Tool Integration - Claude in Excel add-in now supports MCP connectors for tools like S&P Global, LSEG, and Moody's, enabling data pulling without leaving Excel ![Screenshot at 00:07: The comparison table highlighting that Claude Sonnet 4.6 \(59.1% on Agentic terminal coding\) is now matching or outperforming previous models like Opus 4.6 \(65.4%\) and Sonnet 4.5 \(51.0%\) in several agentic benchmarks.](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-00-07.jpg) ![Screenshot at 00:12: Highlighting the Agentic computer use score of 72.5% for Sonnet 4.6, showing improved performance over Sonnet 4.5 \(61.4%\).](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-00-12.jpg) ![Screenshot at 01:33: The 'Money balance over time' graph showing Sonnet 4.6 \(red line\) significantly outpacing Sonnet 4.5 \(grey line\) in the Vending-Bench Arena simulation.](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-01-33.jpg) ![Screenshot at 01:39: The ARC-AGI-2 Leaderboard, a scatter plot comparing model score versus cost per task, visualizing the competitive position of the new models.](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-01-39.jpg) ![Screenshot at 02:54: A tweet from Windsurf announcing that Claude Sonnet 4.6 is live with 1M token context support, available in Arena Mode's Frontier and Hybrid Battle Groups.](https://ss.rapidrecap.app/screens/RttuBBZUvv0/00-02-54.jpg)