Anthropic on Tuesday launched Claude Sonnet 4.6, the latest version of its mainstream AI model, and the new version promises to almost match the company’s flagship, agentic-focused Opus 4.6 that was released two weeks ago, writes Frederic Lardinois, Senior Editor for AI, in today’s lead story.

Here’s how the new Sonnet model compares to its predecessor, Sonnet 4.5, released in September 2025: Sonnet 4.6 easily outperforms it across virtually all standard benchmarks. When it comes to performing office tasks (using OpenAI’s GDPval benchmark), it even beats the latest Opus model, as well as Google’s Gemini 3 Pro and OpenAI’s GPT-5.2.

On coding tasks, where Anthropic has long had an advantage over its competitors, Sonnet 4.6’s benchmark results are often within a percentage point or less of Opus 4.6. Anthropic notes that some of the improvements include better overall consistency and adherence to instructions.

About that “cents” pun: Sonnet 4.6 is also available at a significantly lower price of $3/$15 per million input/output tokens than Opus 4.6, which costs $5/$25 per million input/output tokens.

Go deeper on the new Sonnet 4.6 model here: Anthropic’s new Claude Sonnet 4.6 promises Opus-level coding at Sonnet pricing

Here are more stories worth your brain space today:

➪ The “funhouse mirror”: How AI reflects the hidden truths of your software pipeline

➪ Your AI strategy is built on layers of API sediment

➪ Python virtual environments: isolation without the chaos

Organizations that over-privilege AI systems report dramatically higher incident rates than those enforcing least privilege. Explore the data showing why access scope is the defining AI security control.

As more organizations repatriate, modernize, and build new workloads on-premises, the demand for a “public cloud-like” experience inside the data center has never been higher. Join us on March 12 to explore how to automate full-stack Kubernetes deployments with a single git commit.