What’s New with Claude Sonnet 4.6

• Opus-Level Performance at Sonnet Pricing: Claude Sonnet 4.6 reaches parity with Opus 4.5 on agentic coding and computer-use benchmarks while maintaining the same cost as Sonnet 4.5, delivering significantly more capability without increasing price.
• Massive Context + Smarter Execution: With a 1 million token context window in beta, Sonnet 4.6 can reason across entire codebases or large document collections, showing stronger long-horizon planning, fewer hallucinations, better instruction following, and more consistent multi-step task execution.
• Production-Ready for Real Workflows: Early users report near human-level performance in complex workflows like spreadsheet navigation and multi-tab browser tasks, and integration through DigitalOcean Gradient™ AI Serverless Inference makes it easy to deploy with real-world data and external connectors.

The key focus of this release is that Sonnet 4.6 has reached parity with Opus 4.5, a significantly more expensive model from the last generation, on agentic coding and computer use benchmarks. Additionally, Claude Sonnet 4.6 now supports a 1 million token context window in beta, significantly increasing the value over its predecessor. According to Anthropic, testing in Claude Code showed users preferred Sonnet 4.6 over Sonnet 4.5 about 70% of the time, citing its stronger ability to understand context before editing code, consolidate shared logic instead of duplicating it, and remain reliable over extended sessions. It was even favored over the earlier frontier model, Opus 4.5, 59% of the time, with users reporting less overengineering, fewer signs of “laziness,” better instruction following, fewer hallucinations, and more consistent execution of multi-step tasks.

With a 1M-token context window capable of holding entire codebases or extensive documents in a single prompt, Sonnet 4.6 demonstrates improved long-horizon reasoning, as seen in the Vending-Bench Arena evaluation, where it adopted a strategic approach—investing heavily in capacity early on before pivoting to profitability—ultimately outperforming competing models.

Early users highlight Sonnet 4.6’s ability to reliably complete complex workflows, from spreadsheet operations to multi-tab web navigation. While still not matching top human experts, the rapid pace of progress suggests that practical, broadly capable AI-driven computer use is quickly becoming viable for real-world work. As we can see from the graph above, this steady improvement should soon lead to expert level performance by the AI as it is developed.

So what does this mean in practice for users? First, Sonnet 4.6 combines affordability with frontier-level capability, making it one of the strongest candidates for large-scale deployment across diverse production workloads. With performance now reaching parity with Opus 4.5 in key agentic coding and computer-use benchmarks, teams can confidently substitute Sonnet in scenarios that previously required a higher-cost model without sacrificing reliability or reasoning depth. Second, and perhaps more compelling, its improved computer-use capabilities unlock powerful new automation possibilities. Paired with DigitalOcean’s OpenClaw Droplet, Sonnet 4.6 can function as a highly capable digital operator, handling browser workflows, form submissions, research tasks, and multi-step desktop operations, effectively acting as a scalable, intelligent assistant for both professional and personal productivity.

If you want to use Sonnet 4.6 beyond Claude Code, the most seamless path is through Gradient™ AI Serverless Inference. For most developers, Serverless Inference is the fastest and most flexible choice, enabling you to integrate the model into existing applications typically with just a few lines of code. Create a model access key right now by clicking on the link here to generate a key in the DigitalOcean Cloud Console.

Deploying on DigitalOcean also means combining Sonnet 4.6’s advanced reasoning capabilities with the broader DigitalOcean ecosystem. You can enrich agents with real-world context by attaching custom knowledge bases or connecting external data sources. Built-in connectors simplify data ingestion from services like AWS S3, Dropbox, and other storage providers, eliminating the need for complex configuration while making it easy to operationalize AI in production workflows.

If you want to compare Anthropic models before integrating them into your application, you can use the Gradient™ AI Model Playground to evaluate them side by side.

Sonnet 4.6 marks a meaningful inflection point for the Claude ecosystem. By delivering Opus-level performance at Sonnet pricing, dramatically expanding context capacity to 1 million tokens (beta), and improving real-world reliability in coding and computer-use tasks, Anthropic has strengthened its position in agentic AI workflows almost overnight. What stands out most is not just benchmark parity, but the reported practical usability gains: better instruction following, fewer hallucinations, stronger long-horizon reasoning, and more consistent execution across complex tasks. Combined with seamless deployment through DigitalOcean Gradient™ AI, Sonnet 4.6 feels less like a routine model update and more like a foundational upgrade for developers building production-ready AI systems.