Claude Sonnet 4.6 Achieves 72.5% OSWorld, Matching Opus Computer Use

Computer Use Democratized

Claude Sonnet 4.6's 72.5% score on OSWorld-Verified essentially ties Opus 4.6's 72.7%—bringing sophisticated desktop automation to mid-tier pricing.

What is OSWorld?

OSWorld tests AI models on real computer tasks:

• Web browsing and form filling

• Desktop application use

• File management

• Multi-step workflows

• Cross-application tasks

Performance Comparison

Model

OSWorld-Verified

Price

Opus 4.6

72.7%

$15/$75

Sonnet 4.6

72.5%

$3/$15

Sonnet 4.5

61.4%

$3/$15

GPT-5.2

~65%

$1.75/$14

Sonnet 4.6 improved 11+ points from Sonnet 4.5, reaching Opus parity.

Practical Capabilities

Sonnet 4.6 can now reliably:

Web Automation

• Fill complex forms with validation

• Navigate multi-step checkout flows

• Extract data from dynamic websites

Desktop Tasks

• Manipulate spreadsheets

• Process documents across applications

• Manage file systems

Enterprise Workflows

• Expense report submission

• Data entry automation

• Testing and QA scenarios

Enterprise Interest

RPA vendors are taking notice:

"We're evaluating Sonnet 4.6 for our automation platform. At these performance levels and this price point, AI-first RPA becomes viable for mid-market companies." — VP Product, Automation startup

Implementation Example

python
# Simple form automation with Sonnet 4.6
response = client.messages.create(
    model="claude-sonnet-4-6-20260217",
    max_tokens=4096,
    tools=[{"type": "computer_20241022", "name": "computer", ...}],
    messages=[{
        "role": "user",
        "content": [
            {"type": "image", "source": screenshot},
            {"type": "text", "text": "Fill out this expense form: Date 2/17, Amount $145.50, Category: Travel"}
        ]
    }]
)


Safety Considerations

Anthropic emphasizes safety with computer use:
Sandboxed execution recommended
Human approval for sensitive actions
Audit logging for compliance
Rate limiting to prevent runaway agents
The Pricing Impact

A typical enterprise computer use deployment:
Opus 4.6: ~$1,500/month for 20K tasks
Sonnet 4.6: ~$300/month for same tasks
80% cost reduction with equivalent performance.

What's Next

As computer use matures, expect:
Integration with enterprise RPA platforms
Compliance certifications for regulated industries
More sophisticated multi-step orchestration
Better handling of dynamic/animated UIs
Conclusion

Sonnet 4.6 has removed the cost barrier to AI-powered computer automation. What was premium capability six months ago is now standard tier.