Here is the latest version of the new Claude Sonnet 4.6! Key points:
First — the price has dropped by approximately 40% compared to Opus, and this model is now available for all subscription plans, including the free tier. This applies to platforms like Claude Code, Cowork, and others.
Second — in beta, the model can process context up to one million tokens! This significantly expands capabilities for working with large volumes of data.
Additionally, developers have seriously updated the internal coding, improved reranking processes, enhanced computer utilization, and even achieved an unexpected upgrade in the financial sphere. The model has become more advanced across almost all fronts.
According to benchmark results, Sonnet 4.6 closely rivals its predecessor Opus 4.6 and even outperforms GPT-5.2 and Gemini 3 Pro in many tests — indicating truly high performance.
It’s especially worth noting the substantial improvement in handling computer tasks: the agent now manages large tasks — such as complex spreadsheets or multi-step forms — as well as a human.
Another interesting point — in the Vending Bench test, the model showed excellent results: it independently developed a strategy and earned about $5,639. In comparison, Opus 4.6 managed to earn only around $4,000. However, organizers emphasize that Sonnet employs quite aggressive strategies.
So, this update promises to be quite impressive!
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby