Near-flagship quality at lower cost
Sonnet 4.6 is often the better price-performance option for daily engineering and operations workflows.
Claude Sonnet 4.6 — Anthropic's Strongest Sonnet
Anthropic's most capable Sonnet model. Near Opus-level intelligence with dramatic improvements in coding, computer use, long-context reasoning, and agent planning — at unchanged Sonnet pricing.
OSWorld Score
72.5%
Context Window
1M tokens
Claude Code Preference
70%
Pricing
$3 / $15
Try It Now
Try Claude Sonnet 4.6
Enter a prompt below to experience Claude Sonnet 4.6's dramatically improved capabilities firsthand.
5 credits per message
Sign in to use
Capabilities
What Claude Sonnet 4.6 Can Do
Dramatically improved coding, computer use that approaches human-level performance, and powerful long-context reasoning — all at Sonnet pricing.
claude-sonnet-4-6
Agentic Coding
Reads context before modifying code, consolidates duplicated logic, reduces over-engineering and lazy behavior. Multi-step task execution is far more stable with fewer false-positive success reports.
Computer Use
Evolved from experimental to practical. OSWorld benchmark jumped from 14.9% to 72.5%. Approaches human-level performance on complex spreadsheets and multi-step web forms, coordinating across multiple browser tabs.
Long-Context Reasoning
1M token context window (beta) holds entire codebases, long contracts, or dozens of research papers. Sonnet 4.6 reasons effectively over such long contexts — not just stuffs text in.
Benchmarks
Benchmark Performance
All-around improvement approaching Opus-level performance, with standout results in computer use and coding.
Computer Use
72.5%
OSWorld
Up from 14.9% at launch — approaching human-level performance in real software environments
Coding
70%
Claude Code Preference
70% of Claude Code users prefer Sonnet 4.6 over previous Sonnet 4.5
Coding
59%
vs Opus 4.5
59% of users prefer Sonnet 4.6 over flagship Opus 4.5 for coding tasks. Learn more about <a href='https://www.anthropic.com/models/opus'>Opus</a>.
Agentic Coding
#1
Terminal-Bench 2.0
Highest score on the agentic coding evaluation
Long Context
1M
Context Window
1 million token context window (beta) for entire codebases and long documents
Value
Top Value
Pricing
Best-in-class value for a model that delivers near Opus-level performance
Key Features
Key Features of Claude Sonnet 4.6
A comprehensive upgrade across every dimension — from programming to planning, from safety to scale.
Near Opus-Level Intelligence
Comprehensive benchmark gains across all evaluations. Tasks that previously required Opus can now be handled by Sonnet — at Sonnet pricing of $3/$15 per million tokens.
1M Token Context (Beta)
Enough to hold an entire codebase, a long contract, or dozens of research papers. Sonnet 4.6 reasons effectively over long contexts, enabling complex long-horizon planning.
Improved Prompt Injection Defense
Significant improvement in resisting prompt injection attacks compared to Sonnet 4.5. Malicious webpages can no longer easily hijack the model during computer use tasks.
Agent Planning
Dramatically improved multi-step task execution. Better at orchestrating agent teams, planning ahead, and recovering from errors without human intervention.
Enhanced Web Tools
Web search and scraping tools now automatically filter and process results, keeping only relevant content to save tokens. Code execution, memory, and tool use have reached GA.
Unchanged Pricing
Despite all improvements, pricing remains $3 per million input tokens and $15 per million output tokens — the same as previous Sonnet models.
Why Use
Why choose Claude Sonnet 4.6
A practical balance of strong coding quality, computer-use automation, and predictable cost efficiency.
Strong computer-use automation
It handles multi-step browser and office-like tasks with better reliability and fewer intervention loops.
Reliable long-context reasoning
With 1M context support, teams can process bigger artifacts and preserve task continuity across long sessions.
How to Use
How to get better results with Sonnet 4.6
Use an execution-first prompt structure to improve consistency across coding, automation, and long-context tasks.
Define the workflow before asking
Specify inputs, expected outputs, and failure handling to reduce ambiguity in multi-step automation tasks.
Segment long-context materials
Group documents into sections and mark source priorities so Sonnet can maintain stronger reasoning focus.
Measure cost and quality together
Track token usage, latency, and task completion quality to decide where Sonnet should replace higher-cost models.
Related models
Compare other frontier models on Fluxchat.
GPT-5.4
OpenAI's professional model for coding, reasoning, tools, and long-context work
Claude Opus 4.6
Anthropic's most intelligent model with 1M context
GLM-5
Zhipu AI's flagship model with open SOTA coding & agent performance
Gemini 3.1 Pro
Google's most advanced reasoning model with 77.1% ARC-AGI-2 & 1M context
Frequently Asked Questions
Have more questions? Contact our support team.
Experience Claude Sonnet 4.6 on Fluxchat
Try Anthropic's strongest Sonnet model for coding, computer use, research, and everyday work — right here on Fluxchat.