Claude Sonnet 4.6 Complete Guide: Features, Benchmarks & Pricing
Everything you need to know about Claude Sonnet 4.6: 1M token context, 79.6% SWE-bench score, Adaptive Thinking, and Opus-level performance at Sonnet pricing.
TL;DR
Claude Sonnet 4.6, released February 17, 2026, delivers near-Opus performance at 1/5th the cost. Key highlights: 79.6% SWE-bench Verified, 72.5% OSWorld, 1M token context window (beta), and Adaptive Thinking for dynamic reasoning. Pricing stays at $3/$15 per million tokens—same as Sonnet 4.5.
Release Overview
Anthropic released Claude Sonnet 4.6 on February 17, 2026—just 12 days after launching Opus 4.6. The model is now the default for Free and Pro plan users on Claude.ai, and is available across Claude Code, Claude Cowork, the API, and all major cloud platforms (AWS Bedrock, Google Vertex AI, Microsoft Foundry).
Key Features
1M Token Context Window (Beta)
Sonnet 4.6 supports a 1M token context window—roughly 750,000 words or 5-10 full codebases. This is paired with a new context compaction feature that automatically summarizes older context when approaching the limit, enabling effectively unlimited conversations.
Adaptive Thinking Engine
Replacing the old binary "extended thinking" mode, Adaptive Thinking allows the model to dynamically decide how much reasoning is required. Using the new effort parameter, developers can balance speed, cost, and intelligence based on task complexity.
Enhanced Computer Use
Sonnet 4.6 scores 72.5% on OSWorld-Verified—within 0.2% of Opus 4.6's 72.7%. This makes it the most capable mid-tier model for agentic computer tasks, including web browsing, form filling, and desktop automation.
Improved Web Search
Through the API, Claude's web search generates and runs code to sort and refine search results, retaining only relevant information. This improves response quality while reducing unnecessary token usage.
Benchmark Performance
| Benchmark | Sonnet 4.6 | Opus 4.6 | Sonnet 4.5 |
|---|
| SWE-bench Verified | 79.6% | 80.8% | 77.2% |
| OSWorld-Verified | 72.5% | 72.7% | 61.4% |
| GPQA Diamond | 74.1% | 91.3% | 65.0% |
| Math (AIME) | 89% | 93% | 62% |
| GDPval-AA (Office Tasks) | 1633 Elo | 1606 Elo | - |
Notably, Sonnet 4.6 outperforms Opus 4.6 on real-world office tasks (GDPval-AA) and agentic financial analysis (63.3% vs 60.1%).
User Preferences
In early testing:
- Users preferred Sonnet 4.6 over Sonnet 4.5 70% of the time
- Users preferred Sonnet 4.6 over Opus 4.5 59% of the time
- Most coding tasks (79.6% SWE-bench)
- Computer use and automation
- Office/knowledge work tasks
- Financial analysis
- Cost-sensitive applications
- Expert-level science/research reasoning (91.3% GPQA)
- Multi-agent coordinated tasks
- Maximum needle-in-haystack retrieval (76% vs 18.5%)
Testers cited better instruction following, fewer hallucinations, and reduced over-engineering as key improvements.
Pricing
| Model | Input ($/M) | Output ($/M) |
|---|
| Claude Sonnet 4.6 | $3 | $15 |
| Claude Opus 4.6 | $15 | $75 |
Sonnet 4.6 delivers ~98-99% of Opus performance at 20% of the cost. Additional savings available: 90% with prompt caching, 50% with batch processing.
When to Use Sonnet 4.6 vs Opus 4.6
Use Sonnet 4.6 for:
Escalate to Opus 4.6 for:
Getting Started
import anthropicclient = anthropic.Anthropic()
response = client.messages.create(
model="claude-sonnet-4-6-20260217",
max_tokens=8192,
messages=[{"role": "user", "content": "Your prompt here"}]
)
Conclusion
Claude Sonnet 4.6 represents the new sweet spot in AI: flagship-class performance at mid-tier pricing. For most developers, it should be the default choice, with Opus reserved for edge cases requiring maximum reasoning depth.