Claude Sonnet 4.6 Review: Benchmarks, Pricing & What's New
Anthropic's newest mid-tier model brings 1M token context, major computer use improvements, and coding quality that beats Opus 4.5 — at the same $3/1M price as Sonnet 4.5.
Claude Sonnet 4.6 at a Glance
Input Price
$3.00 / 1M tokens
Output Price
$15.00 / 1M tokens
Context Window
1M tokens (beta)
Released
Feb 17, 2026
Claude Sonnet 4.6, released February 17, 2026, is Anthropic's most significant mid-tier model upgrade to date. It's now the default model on claude.ai for Free and Pro users — replacing Sonnet 4.5 — and brings a 1M token context window in beta, dramatically improved computer use capabilities, and coding quality that surpasses the previous Opus-class model (Claude Opus 4.5) in real-world developer preference testing.
What's New in Claude Sonnet 4.6
Coding Improvements
- • More effective context reading before modifying code
- • Consolidates shared logic instead of duplicating it
- • Fewer hallucinations and false claims of success
- • Less prone to overengineering and "laziness"
- • Better instruction following over long sessions
Computer Use
- • Major improvement on OSWorld-Verified benchmark
- • Human-level capability on spreadsheets and web forms
- • Multi-tab browser workflows now reliable
- • 16 months of steady OSWorld benchmark gains
- • Major improvement in prompt injection resistance
Context Window
- • 1M token context window (beta) — first for Sonnet-class
- • Up from 200K tokens in Sonnet 4
- • Matches Opus 4.6's context at a lower price
- • Suitable for full repository ingestion
Safety & Character
- • "Broadly warm, honest, prosocial, and at times funny"
- • Very strong safety behaviors per system card
- • No signs of major misalignment concerns
- • Improved prompt injection resistance vs Sonnet 4.5
- • Similar prompt injection safety to Opus 4.6
Coding Performance: Preference Data
Anthropic ran extensive preference testing in Claude Code — their agentic coding environment — before releasing Sonnet 4.6. The results show significant improvements over both Sonnet 4.5 and the previous Opus-class frontier:
Users preferred Sonnet 4.6 over Sonnet 4.5 70% of the time in Claude Code. They rated it as significantly less prone to overengineering and "laziness," with better instruction following.
Users even preferred Sonnet 4.6 to Opus 4.5 — Anthropic's frontier from November 2025 — 59% of the time. The key reasons: fewer false claims of success, fewer hallucinations, and more consistent multi-step follow-through.
This is a significant result. Sonnet 4.6 costs $3/1M input tokens compared to Opus 4.5's $4/1M — meaning it's cheaper AND preferred for everyday coding tasks. "Performance that would have previously required reaching for an Opus-class model is now available with Sonnet 4.6," according to Anthropic's release.
Computer Use: The Benchmark Story
Computer use — the ability for Claude to control a computer via mouse clicks and keyboard input — has been one of Anthropic's most actively improved capabilities. Sonnet 4.6 shows a major leap on OSWorld-Verified, the standard benchmark for AI computer use. OSWorld presents hundreds of tasks across real software (Chrome, LibreOffice, VS Code) on a simulated computer with no special APIs.
What users are seeing in production:
- Spreadsheet navigation: Human-level capability on complex, multi-sheet spreadsheets
- Multi-step web forms: Reliable form completion across multiple tabs and pages
- Legacy software automation: Interacting with systems that have no modern APIs
- Prompt injection resistance: Major improvement vs Sonnet 4.5; similar protection to Opus 4.6
Pricing: Same Cost, Much More Capability
Claude Sonnet 4.6 pricing is unchanged from Sonnet 4.5: $3 per million input tokens and $15 per million output tokens. This is a significant value improvement — you get a substantially more capable model for the same spend.
| Model | Input (1M tokens) | Output (1M tokens) | Context Window | Best For |
|---|---|---|---|---|
| Claude Sonnet 4.6 ✦ New | $3.00 | $15.00 | 1M (beta) | Coding, computer use, agents |
| Claude Opus 4.6 | $5.00 | $25.00 | 1M (beta) | Complex agentic tasks, finance, legal |
| Claude Sonnet 4.5 | $3.00 | $15.00 | 200K | Previous default (upgraded to Sonnet 4.6) |
| GPT-5.2 | $2.50 | $10.00 | 256K | Reasoning, mathematical coding |
| Claude Sonnet 4 (Oct 2025) | $1.00 | $5.00 | 200K | Budget-conscious production |
Prices per million tokens as of February 6, 2026.
Sonnet 4.6 vs Opus 4.6: When to Use Each
With Sonnet 4.6 now outperforming Opus 4.5 in user preference tests, the question becomes: when do you still need Opus 4.6? According to Anthropic, the answer is the most demanding tasks — where Opus 4.6 leads all frontier models:
Use Claude Sonnet 4.6 for:
- • Everyday coding, debugging, and code review
- • Computer use tasks (forms, spreadsheets, legacy software)
- • Long-context document processing (up to 1M tokens)
- • Agent workflows requiring good judgment
- • Production apps where cost × quality matters
- • Default model on claude.ai (Free + Pro)
Use Claude Opus 4.6 for:
- • Maximum accuracy tasks: Humanity's Last Exam, GPQA Diamond
- • Finance & legal analysis: GDPval-AA leader (144 Elo ahead of GPT-5.2)
- • Multi-step agentic coding: Terminal-Bench 2.0 #1 score
- • Hard research tasks: BrowseComp leader for hard-to-find info
- • Long-horizon autonomous tasks requiring deep planning
For most developers and teams, Sonnet 4.6 will cover 80–90% of use cases at 40% lower cost than Opus 4.6. Reserve Opus 4.6 for tasks where maximum accuracy on complex, multi-step reasoning is required.
Test Claude Sonnet 4.6 vs other models
Want more detailed comparisons with scoring and benchmarks?
Who Should Upgrade to Sonnet 4.6?
Upgrade immediately if you:
- • Use Claude for coding and software development
- • Need computer use capabilities for automation
- • Are currently on Sonnet 4.5 (same price, much better)
- • Want the 1M token context window for large documents
Evaluate carefully if you:
- • Currently use Claude Sonnet 4 (Oct 2025) for $1/1M input
- — Sonnet 4.6 at $3/1M is 3× more expensive; test whether the quality improvement justifies cost for your workload
- • Have highly customized system prompts tuned to Sonnet 4.5 behavior
Stick with Opus 4.6 if you:
- • Run complex financial or legal document analysis
- • Need maximum benchmark accuracy (Humanity's Last Exam, GPQA)
- • Use long-horizon agentic workflows requiring deep planning
Frequently Asked Questions
Is Claude Sonnet 4.6 better than Sonnet 4.5?
Yes, significantly. According to Anthropic, users preferred Sonnet 4.6 over Sonnet 4.5 70% of the time in Claude Code testing. Key improvements include better coding consistency, 1M token context (vs 200K), major computer use gains on OSWorld-Verified, and stronger prompt injection resistance. Pricing is unchanged at $3/$15 per 1M tokens.
What does Claude Sonnet 4.6 cost?
Claude Sonnet 4.6 costs $3 per million input tokens and $15 per million output tokens via the Anthropic API — the same pricing as Sonnet 4.5. It is the default model on claude.ai's Free and Pro plans. The API model ID is claude-sonnet-4-6.
Does Claude Sonnet 4.6 have a 1M token context?
Yes. Claude Sonnet 4.6 includes a 1M token context window in beta, up from 200K in Sonnet 4. This is the first time a Sonnet-class model has matched Opus-class context, and it opens up use cases like full repository ingestion and very large document processing.
Should I use Sonnet 4.6 or Opus 4.6?
For most tasks — coding, computer use, long-context reasoning, knowledge work — Sonnet 4.6 is the better choice at $3/1M (vs $5/1M for Opus 4.6). Anthropic's own testing showed Sonnet 4.6 is preferred over the previous Opus-class model 59% of the time. Use Opus 4.6 for tasks requiring maximum accuracy: complex financial analysis, research requiring Humanity's Last Exam-level reasoning, or long-horizon agentic coding where Terminal-Bench performance matters.
Is Claude Sonnet 4.6 available on the free plan?
Yes. As of February 17, 2026, Claude Sonnet 4.6 is the default model on claude.ai for both Free and Pro users. It's also available via the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and all major cloud platforms.
Conclusion
Claude Sonnet 4.6 is the most significant mid-tier model upgrade Anthropic has shipped. It delivers Opus 4.5-level performance — and beats Opus 4.5 in user preference — at the same $3/1M price as Sonnet 4.5. The 1M token context window puts it on par with Opus 4.6 for most long-context use cases. The computer use improvements on OSWorld-Verified represent 16 months of compounded progress.
For the majority of developers and teams, Sonnet 4.6 is now the right default choice. Save Opus 4.6 for the tasks where frontier intelligence — GDPval-AA, Terminal-Bench, Humanity's Last Exam — genuinely matters.
Compare Claude Sonnet 4.6 vs Other Models
Test Sonnet 4.6, Opus 4.6, GPT-5.2 and more with your own prompts. Free, no signup required.