Anthropic's new model, Claude Sonnet 4.6, was unveiled on February 17, 2026, with a clear message: this is the best AI model on the market for navigating and performing tasks on a PC without human intervention. According to digi.no, the company is explicitly positioning the model for autonomous PC use, meaning the model can control interfaces, run programs, and complete tasks on its own.
Strong Performance on Intelligence Tests
Independent benchmarks from the Artificial Analysis Intelligence Index support parts of Anthropic's claims. In adaptive reasoning mode with maximum effort, Claude Sonnet 4.6 achieves an Intelligence Index score of 51, well above the median of 26 for comparable models.
On GPQA — a benchmark testing PhD-level reasoning in natural sciences — the model scores 87.5 percent in adaptive mode. This is marginally lower than OpenAI's GPT-5.2, which according to leaderboards from Artificial Analysis achieves 90.3 percent on certain subtests, but still among the very highest scores in the field.
Competition is Tight — and Cheap
Competition for the top of the AI market is fiercer than it has been in a long time. While Anthropic and OpenAI duel over fractions on demanding benchmarks, Chinese DeepSeek has caused headaches for American players — not by matching them on peak performance, but by offering impressive results at a fraction of the price.
This price pressure forces established players to justify their premium pricing more clearly, which may explain Anthropic's distinct focus on a concrete use case: autonomous PC control. This is an area where Chinese competitors have not yet presented equally specific claims.
However, there is a significant weakness in the data: there is a lack of published, independent benchmark results specifically for "computer use" — i.e., agentic tasks such as GUI navigation and terminal control. Anthropic's claim of market leadership in this area is currently largely the company's own word, according to available market data.
Speed is an Achilles' Heel
One weakness emerging in independent measurements is generation speed. Claude Sonnet 4.6 produces between 56 and 57 tokens per second, which is below the median of 60.5 and 71.2 tokens per second respectively, depending on the mode. For applications where response time is critical, this could be an argument for competitors.
Anthropic claims market leadership in autonomous PC control — but independent benchmarks for this specific use case are currently absent
A Growing Market
The launch takes place in a rapidly growing market. Estimates for the AI agent and process automation market vary significantly between analysis firms but consistently point upward with annual growth exceeding 40 percent. The market value for agentic AI is estimated by various sources to be between $4.5 and over $15 billion in 2025, depending on how the segment is defined.
For 2026, the market for AI agents alone is estimated to pass $11 billion, according to industry reports. This underscores why players are fighting so hard to define who leads in autonomous task execution — that is where the money is flowing.
What Happens Next?
Claude Sonnet 4.6 is available via Anthropic's API at $3 per million input tokens and $15 per million output tokens. The model represents a clear strategic choice by Anthropic to differentiate itself through agentic use rather than pure text generation or conversation — a race where we can expect more major moves from all players during 2026.