TL;DR
- Anthropic just launched Claude Sonnet 4.6. Their most capable Sonnet model ever
- It now comes with a 1M token context window (enough to load entire codebases in one go)
- Computer use skills have improved dramatically which is closer to human-level for real tasks
- Free users now get Sonnet 4.6 by default on claude.ai
- Pricing stays the same: $3/$15 per million tokens
- In head-to-head tests, users preferred Sonnet 4.6 over the older Opus 4.5 59% of the time
- Stacks up well against GPT-5.2 and Gemini 3 Pro across benchmarks
If you’ve been using Claude for writing, coding or research. This week just got more interesting. Anthropic quietly dropped Claude Sonnet 4.6 and based on what’s under the hood, it’s not a small update. It’s the kind of release that makes you rethink which AI tool deserves a spot in your daily workflow.
Here’s everything you need to know.
What Is Claude Sonnet 4.6?
Claude Sonnet 4.6 is Anthropic’s latest mid-tier model. Sitting between the everyday Claude Haiku and the heavyweight Opus line. But mid-tier undersells it this time around. Anthropic describes it as their most capable Sonnet model yet. With improvements across coding, long-context reasoning, computer use and design tasks.
What makes this launch stand out is that Sonnet 4.6 is now the default model for all Claude users. Including the free plan. You don’t need to upgrade to experience it. It’s already there when you open claude.ai.
What’s Actually New?
1M Token Context Window
Sonnet- 4.6 ships with a 1 million token context window in beta. To put that in perspective. You can paste in an entire software codebase, a stack of research papers, or months of financial records and the model processes all of it in a single request. More impressively, it doesn’t just store that context. It reasons across it. That’s a meaningful difference.
Computer Use Gets Seriously Better
Back in October 2024, Anthropic was first to launch a general-purpose computer-using AI model. They admitted at the time it was experimental and clunky. Sonnet 4.6 is the version where it starts to feel real. Early users are seeing near human-level performance on tasks like navigating spreadsheets, filling out multi-step web forms and managing workflows across multiple browser tabs.All without custom connectors or special APIs.
Coding That Rivals Opus
In Claude Code testing, users preferred Sonnet- 4.6 over the previous Sonnet 4.5 roughly 70% of the time. They even preferred it over Opus 4.5 Anthropic’s previous flagship 59% of the time. The feedback? Less overengineering, fewer hallucinations and better follow-through on complex multi-step tasks.
Design and Frontend Polish
This one surprised early testers. Customers independently described visual outputs from Sonnet 4.6 as noticeably more polished. Better layouts, smoother animations, stronger design instincts. One team said it reached for modern tooling they didn’t even ask for and delivered production-ready results in one shot.
Sonnet 4.6 vs Sonnet 4.5
| Feature | Claude Sonnet 4.5 | Claude Sonnet 4.6 |
|---|---|---|
| Context Window | 200K tokens | 1M tokens (beta) |
| Computer Use | Basic, experimental | Near human-level on tasks |
| Coding Preference | Baseline | Preferred 70% over 4.5 |
| Pricing | $3/$15 per million tokens | Same — $3/$15 per million tokens |
| Default on Free Plan | No | Yes |
| Extended Thinking | Yes | Yes + Adaptive Thinking |
| Prompt Injection Resistance | Moderate | Major improvement |
| Design Output Quality | Standard | Noticeably more polished |
You might be interested in Claude cowork
How Does It Stack Up Against the Competition?
This is where it gets genuinely interesting for anyone who has been comparing AI tools.
Sonnet 4.6 vs GPT-5.2: Sonnet matches or outperforms GPT-5.2 on computer use benchmarks. A category where OpenAI has historically been strong. On real-world office tasks. Sonnet new model delivers Opus-level performance. Which is a tier above what GPT-5.2 reaches at a comparable price point.
Sonnet 4.6 vs Gemini 3 Pro : Google’s Gemini 3 Pro is a capable model, but Sonnet 1M context window and agentic planning capabilities give it a practical edge for long-horizon tasks. The kind that involve multiple steps, multiple tools and sustained reasoning over time. Gemini’s strength remains multimodal tasks but for document reasoning and code. Sonnet 4.6 holds its ground.
The bottom line: At $3/$15 per million tokens, Sonnet 4.6 offers frontier-level results without frontier-level pricing. That performance-to-cost ratio is hard to beat right now.
Who Should Care Most
- Developers building agentic apps or managing large codebases
- Content creators using AI for research, drafting, and long-form writing
- Businesses processing enterprise documents, contracts or financial reports
- Free Claude users — you already have access, no upgrade needed
FAQ
Is Claude Sonnet 4.6 free?
Yes. It’s now the default model on Anthropic’s free plan at claude.ai. No subscription required to try it.
How is Sonnet 4.6 different from Claude Opus?
Opus 4.6 is still the stronger choice for the deepest reasoning tasks — codebase refactoring, coordinating multiple AI agents and problems where precision is non-negotiable. But Sonnet 4.6 closes that gap significantly, at a fraction of the cost.
Can Sonnet 4.6 really use a computer?
Yes and meaningfully better than before. It can click, type, navigate browsers, and fill forms the same way a person would, without needing custom integrations. It still lags behind the most skilled humans, but the progress over 16 months has been remarkable.
Is the 1M token context window available now?
It’s available in beta right now via the API. Full rollout is expected to follow.
Sources
- Anthropic – Official Claude Sonnet 4.6 Announcement: https://www.anthropic.com/news/claude-sonnet-4-6
- Anthropic – Claude Model Overview: https://www.anthropic.com/claude
- OSWorld Benchmark – AI Computer Use Evaluation: https://os-world.github.io
- Anthropic API Docs – Prompt Injection Safety: https://docs.anthropic.com/en/docs/computer-use
