Claude Sonnet 4.6: Anthropic's Most Capable Mid-Tier Model Yet

Anthropic just dropped Claude Sonnet 4.6, and it’s not just an incremental update—it’s a fundamental shift in what mid-tier AI models can do.

Here’s the headline: Sonnet 4.6 delivers Opus-level performance at Sonnet pricing. That’s not marketing speak. Early users prefer it to Opus 4.5 (Anthropic’s flagship from November 2025) 59% of the time. And it costs 40% less.

If you’ve been rationing your Opus API calls because of cost, or settling for “good enough” results from cheaper models, this changes the game.

What Makes Sonnet 4.6 Different?

1. It’s Smarter Than the Previous Flagship

Let’s start with the most surprising stat: In Claude Code, users preferred Sonnet 4.6 over Opus 4.5 59% of the time.

That’s not a typo. The new mid-tier model beats the old flagship model in real-world developer workflows.

Why?

Less overengineering: Opus 4.5 had a tendency to overcomplicate solutions. Sonnet 4.6 is more pragmatic.
Better instruction following: It does what you ask, not what it thinks you meant.
Fewer false claims: When it says it’s done, it’s actually done.
More consistent multi-step execution: Long tasks don’t drift off course.

When compared to its direct predecessor (Sonnet 4.5), the preference jumps to 70%. That’s a landslide.

2. Computer Use Just Got Real

In October 2024, Anthropic introduced the first general-purpose AI that could use a computer like a human: clicking, typing, navigating software without APIs.

At the time, it was “experimental—at times cumbersome and error-prone.”

Sixteen months later, Sonnet 4.6 has reached human-level capability on many computer use tasks.

What does “computer use” actually mean?

The model can:

Navigate complex spreadsheets
Fill out multi-step web forms
Switch between browser tabs to gather information
Use software that has no API (legacy systems, specialized tools)
Interact with real applications: Chrome, LibreOffice, VS Code, etc.

Why does this matter?

Most organizations have software they can’t easily automate. Internal tools built before modern APIs existed. Specialized systems with no integration options. Previously, you’d need to build custom connectors for each one.

Now, you can just point Claude at the screen and tell it what to do.

The benchmark numbers:

On OSWorld (the standard benchmark for AI computer use), Sonnet 4.6 scores significantly higher than any previous Sonnet model. It’s not perfect—it still lags behind the most skilled humans—but the rate of improvement is remarkable.

And crucially, Anthropic has made major strides in prompt injection resistance. Computer use models are vulnerable to malicious instructions hidden on websites. Sonnet 4.6 is much harder to hijack than its predecessor.

3. 1M Token Context That Actually Works

Sonnet 4.6 is the first Sonnet-class model with a 1 million token context window (in beta).

That’s enough to hold:

An entire codebase
Dozens of research papers
Lengthy contracts and legal documents
Hundreds of pages of documentation

But here’s the key: it can actually reason across all that context.

Many models claim large context windows but suffer from “context rot”—performance degrades as conversations get longer. Sonnet 4.6 maintains peak performance even at extreme context lengths.

Real-world example:

In the Vending-Bench Arena evaluation (which simulates running a business over time), Sonnet 4.6 developed a sophisticated strategy:

Invest heavily in capacity for the first 10 months (spending more than competitors)
Pivot sharply to profitability in the final stretch
Finish well ahead of the competition

This required long-horizon planning across hundreds of thousands of tokens. Previous models couldn’t maintain coherence over that span.

4. Opus-Level Performance on Enterprise Tasks

On GDPval-AA (an evaluation of economically valuable office tasks), Sonnet 4.6 approaches Opus-level performance.

On OfficeQA (reading enterprise documents, extracting facts, reasoning from them), it matches Opus 4.6.

Translation: Tasks that previously required the most expensive model can now run on the mid-tier model.

What this means for costs:

If you’re running 1 million API calls per month on Opus 4.6 ($5 input / $25 output per million tokens), switching to Sonnet 4.6 ($3 input / $15 output) saves you 40% with minimal performance loss.

For many use cases, the performance loss is zero—or even negative (Sonnet 4.6 is better).

What Early Users Are Saying

Anthropic shared feedback from early access partners. Here are the highlights:

Coding and Development

“Claude Sonnet 4.6 delivers frontier-level results on complex app builds and bug-fixing. It’s becoming our go-to for the kind of deep codebase work that used to require more expensive models.”

“Claude Sonnet 4.6 produced the best iOS code we’ve tested for Rakuten AI. Better spec compliance, better architecture, and it reached for modern tooling we didn’t ask for, all in one shot.”

“Out of the gate, Claude Sonnet 4.6 is already excelling at complex code fixes, especially when searching across large codebases is essential.”

Computer Use

“Claude Sonnet 4.6 hit 94% on our insurance benchmark, making it the highest-performing model we’ve tested for computer use. This kind of accuracy is mission-critical to workflows like submission intake and first notice of loss.”

“We’ve been impressed by how accurately Claude Sonnet 4.6 handles complex computer use. It’s a clear improvement over anything else we’ve tested in our evals.”

Enterprise Workflows

“Box evaluated how Claude Sonnet 4.6 performs when tested on deep reasoning and complex agentic tasks across real enterprise documents. It demonstrated significant improvements, outperforming Claude Sonnet 4.5 in heavy reasoning Q&A by 15 percentage points.”

“Claude Sonnet 4.6 meaningfully improves the answer retrieval behind our core product—we saw a significant jump in answer match rate compared to Sonnet 4.5 in our Financial Services Benchmark.”

Design and Frontend

“Claude Sonnet 4.6 has perfect design taste when building frontend pages and data reports, and it requires far less hand-holding to get there than anything we’ve tested before.”

“Customers independently described visual outputs from Sonnet 4.6 as notably more polished, with better layouts, animations, and design sensibility than those from previous models.”

Technical Improvements

Adaptive Thinking

Previously, you had a binary choice: enable extended thinking or don’t.

Now, with adaptive thinking, Claude decides when deeper reasoning would be helpful. You can adjust the effort level (low, medium, high, max) to control how selective it is.

This gives you fine-grained control over the intelligence/speed/cost tradeoff.

Context Compaction (Beta)

Long-running conversations often hit the context window limit.

Context compaction automatically summarizes and replaces older context when the conversation approaches a threshold. This lets Claude perform longer tasks without hitting limits.

Web Search with Dynamic Filtering

Claude’s web search and fetch tools now automatically write and execute code to filter and process search results, keeping only relevant content in context.

This improves both response quality and token efficiency.

MCP Connectors in Excel

If you use Claude in Excel, you can now connect to external tools like:

S&P Global
LSEG
Daloopa
PitchBook
Moody’s
FactSet

Claude can pull in context from outside your spreadsheet without you leaving Excel.

Pricing and Availability

Pricing: $3 input / $15 output per million tokens (same as Sonnet 4.5)

Premium pricing for 1M context: $10 input / $37.50 output per million tokens (for prompts exceeding 200k tokens)

Available on:

Claude.ai (Free and Pro plans)
Claude Cowork
Claude Code
Claude API
All major cloud platforms

Free tier upgrade: The free tier now defaults to Sonnet 4.6 and includes file creation, connectors, skills, and compaction.

When to Use Sonnet 4.6 vs Opus 4.6

Anthropic still recommends Opus 4.6 for tasks that demand the deepest reasoning:

Codebase refactoring: Large-scale architectural changes
Multi-agent coordination: Orchestrating multiple AI agents in a workflow
Mission-critical accuracy: When getting it exactly right is paramount

For everything else, Sonnet 4.6 is now the default choice.

Rule of thumb:

If you’re currently using Opus 4.5 or earlier → try Sonnet 4.6 first
If you’re using Sonnet 4.5 → upgrade to Sonnet 4.6 immediately
If you need Opus 4.6 → you’ll know (and you’re probably already using it)

Safety and Alignment

Anthropic ran extensive safety evaluations on Sonnet 4.6. The conclusion:

“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.”

Key safety improvements:

Low rate of misaligned behaviors (deception, sycophancy, cooperation with misuse)
Lowest over-refusal rate of any recent Claude model (fewer false rejections of benign queries)
Major improvement in prompt injection resistance (especially important for computer use)

The Bigger Picture: AI Model Economics Are Shifting

Sonnet 4.6 represents a broader trend: the performance gap between mid-tier and flagship models is collapsing.

A year ago, you needed the most expensive model for complex tasks. Six months ago, you could get away with mid-tier for some use cases. Today, the mid-tier model beats last quarter’s flagship.

What this means for developers:

Cost optimization is easier: You can run more workloads on cheaper models without sacrificing quality.
Experimentation is cheaper: Testing new ideas doesn’t require expensive API calls.
Scale is more accessible: Running AI at scale is no longer prohibitively expensive.

What this means for enterprises:

ROI improves: The same budget goes further.
More use cases become viable: Tasks that were too expensive at Opus pricing are now practical.
Competitive advantage shifts: It’s no longer about who can afford the best model—it’s about who can deploy it most effectively.

How to Get Started

For Developers

Use claude-sonnet-4-6 via the Claude API.

Quick start:

import anthropic

client = anthropic.Anthropic(api_key="your-api-key")

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain quantum computing in simple terms"}
    ]
)

print(response.content)

Recommended settings:

Start with default effort (high)
Enable adaptive thinking for complex tasks
Use context compaction for long-running conversations
Experiment with effort levels to find the right balance

For Claude.ai Users

Sonnet 4.6 is now the default model on Free and Pro plans. Just start a new conversation—you’re already using it.

For Enterprise Teams

If you’re on Team or Enterprise plans:

Sonnet 4.6 is available in Claude Cowork and Claude Code
MCP connectors work in Claude in Excel
Contact your account manager for migration guidance

Migration Tips

If you’re currently using Sonnet 4.5 or Opus 4.5:

Test on representative tasks: Run your typical workloads through Sonnet 4.6 and compare results.
Adjust effort levels: Sonnet 4.6 performs well even with extended thinking off. Experiment to find the optimal balance.
Monitor token usage: Context compaction can reduce token consumption on long conversations.
Update prompts: Sonnet 4.6 follows instructions better, so you may be able to simplify your prompts.
Measure cost savings: Track your API spend before and after migration.

The Bottom Line

Claude Sonnet 4.6 is a rare example of a mid-tier model that genuinely competes with flagship models from just a few months ago.

Key takeaways:

✅ Opus-level performance at Sonnet pricing (40% cost savings)
✅ Human-level computer use on many tasks
✅ 1M token context that actually works
✅ 70% user preference over Sonnet 4.5
✅ 59% user preference over Opus 4.5 in coding tasks
✅ Same pricing as Sonnet 4.5 ($3/$15 per million tokens)

If you’re using Claude for coding, computer use, long-context reasoning, or enterprise workflows, Sonnet 4.6 should be your new default.

And if you’ve been avoiding Claude because of cost, this is your moment to reconsider.

Related Reading:

Try Claude Sonnet 4.6:

Claude.ai (Free tier now includes Sonnet 4.6)
Claude API
Claude Code

Stay updated with the latest AI news at ChatGPT2Notion Blog