Anthropic just launched Claude Sonnet 4.6, and it's a major step forward. In internal testing, users preferred Sonnet 4.6 over Sonnet 4.5 approximately 70% of the time. Even more striking: users favored it over Claude Opus 4.5 (the previous flagship model from November 2025) in 59% of comparisons.
Claude Sonnet 4.6 is available to use right now on Metir AI in both standard and thinking modes, alongside every other leading AI model.
What Makes Claude Sonnet 4.6 Special
Claude Sonnet 4.6 is not an incremental bump. Anthropic describes it as "our most capable Sonnet model yet," with improvements across coding, reasoning, computer use, agent planning, and knowledge work. Here's what changed.
1 Million Token Context Window (Beta)
Sonnet 4.6 introduces a 1 million token context window in beta. That's roughly 750,000 words of input you can feed into a single conversation. This unlocks practical workflows that were previously limited to Opus-tier models:
- Process entire codebases or repositories in one prompt
- Analyze lengthy contracts, legal filings, or research collections
- Maintain deep context across extended multi-step agent workflows
- Navigate large documents without losing track of earlier details
Adaptive Thinking
Claude Sonnet 4.6 supports adaptive thinking, where the model intelligently decides when and how deeply to reason through a problem. Instead of requiring manual configuration with fixed thinking budgets, adaptive thinking picks up contextual cues and automatically adjusts its reasoning depth.
This means you get deep analytical reasoning on hard problems and fast, direct responses on simple ones, without having to toggle anything.
Stronger Coding and Computer Use
Coding was already a strength for Claude Sonnet, and 4.6 pushes it further. Improvements include better instruction following, more consistent outputs, and reduced hallucinations. Developers will notice fewer instances of Claude "going off-script" or introducing unnecessary changes.
On the computer use front, Sonnet 4.6 shows substantial progress on OSWorld benchmarks. Users report human-level capability in tasks like complex spreadsheet navigation and multi-step web form completion across browser tabs.
Improved Safety and Prompt Injection Resistance
Anthropic's safety evaluation found that Sonnet 4.6 exhibits "a broadly warm, honest, prosocial, and at times funny character, very strong safety behaviors, and no signs of major concerns around high-stakes forms of misalignment."
Notably, Sonnet 4.6 shows major improvements in prompt injection resistance compared to Sonnet 4.5, performing similarly to Opus 4.6 on security benchmarks. This matters for production deployments where models process untrusted user input.
Real-World Performance
The benchmarks tell one story. Real-world feedback tells another, and both point in the same direction.
Enterprise Adoption
Organizations like Databricks report that Sonnet 4.6 matches Opus 4.6 performance on their OfficeQA evaluations. Pace achieved 94% accuracy on insurance benchmarks, making it the highest-performing model they've tested for computer use workflows. GitHub and Cognition both report significant practical improvements in autonomous coding tasks.
Developer Experience
Claude Sonnet 4.6 is better at following instructions precisely, producing consistent outputs, and handling complex multi-step tasks without drifting. For developers building AI-powered products, this translates directly to fewer retries, less post-processing, and more reliable automation.
Pricing: Same Cost, Better Performance
Anthropic kept pricing identical to Sonnet 4.5:
| Input | Output | |
|---|---|---|
| Price per million tokens | $3.00 | $15.00 |
This makes Sonnet 4.6 one of the best value propositions in AI right now. You're getting a model that users prefer over the previous Opus flagship, at a fraction of the cost.
Claude Sonnet 4.6 vs Sonnet 4.5: What Changed
| Feature | Claude Sonnet 4.5 | Claude Sonnet 4.6 |
|---|---|---|
| Context Window | 200K | 200K / 1M (beta) |
| Adaptive Thinking | No | Yes |
| User Preference | Baseline | 70% preferred |
| Prompt Injection Resistance | Standard | Major improvement |
| Computer Use | Good | Human-level on many tasks |
| Pricing | $3/$15 per MTok | $3/$15 per MTok |
Try Claude Sonnet 4.6 Free on Metir AI
Claude Sonnet 4.6 is available right now on Metir AI in two configurations:
- Claude Sonnet 4.6 - Fast, direct responses for everyday tasks
- Claude Sonnet 4.6 Thinking - Adaptive deep reasoning for complex problems
On Metir AI, you also get access to Claude Opus 4.6 for maximum intelligence, plus GPT-5.4, Gemini 3 Pro, Grok, Perplexity, and more. No need to manage multiple subscriptions. Choose the right model for each task and switch freely between them.
Whether you need Sonnet 4.6 for its coding precision, Opus 4.6 for deep reasoning, or GPT-5.4 for mathematical analysis, Metir AI gives you unified access to every leading model from a single platform.
Start using Claude Sonnet 4.6 free on Metir AI →
Want to learn more about Claude's flagship model? Read our deep dive: Claude Opus 4.6: Anthropic's Most Intelligent AI Model for Coding & Agents