The video highlights the release of Anthropic’s Sonnet 5 model, praised for its advanced agentic capabilities and orchestration skills similar to Fable 5, but notes its high token consumption and cost compared to other models like GPT-5.5 and Opus. While the speaker remains cautious about extensively using Sonnet 5 due to efficiency concerns, they are optimistic about the future of AI models that combine smart decision-making with effective task management.
The video discusses two major updates exciting for Claude fans: the release of a new model called Sonnet 5 and the unbanning of Fable 5 by the Secretary of Commerce, lifting previous restrictions. The speaker expresses happiness about Fable 5’s return but notes a cautious approach to using Sonnet 5 extensively. The focus then shifts to what Anthropic has shared about Sonnet 5, highlighting its advanced agentic capabilities, such as planning, tool usage (browsers and terminals), and autonomous operation that previously required larger, more expensive models.
Sonnet 5 is positioned as a significant upgrade in Anthropic’s lineup, effectively raising the tier for their models. The speaker explains that tasks previously handled by models like Haiku, Sana, and Opus are now being shifted to Sonnet, Opus, and Fable respectively, which also implies increased costs for users. The introductory pricing for Sonnet 5 is set at $2 per million input tokens and $10 per million output tokens until August 31st, after which prices will rise to match those of the Sauna model. This pricing strategy reflects the model’s premium capabilities but also its higher operational costs.
A notable downside of Sonnet 5 is its inefficiency in token usage. Compared to other models like Opus and GPT-5.5, Sonnet consumes significantly more tokens—almost twice as many as Opus and up to five times more than GPT-5.5 on certain tasks. This inefficiency is not due to poor design but rather because Sonnet is built to persistently work until it finds an answer, even if it’s not the smartest approach. This characteristic makes it more expensive to run, especially if the wrong model version is chosen for a task, highlighting the need for engineers to carefully manage model selection.
What makes Sonnet 5 particularly interesting is its behavior, which resembles that of Fable 5. It excels at using sub-agents and orchestrating tasks by breaking down complex work into smaller parts and managing them effectively. This orchestration ability was a standout feature of Fable 5 and is what made it exciting to the speaker. Sonnet 5’s capability to handle such delegation suggests it could serve as a valuable tool for other, smarter models that better understand orchestration, such as Fable 5, Mythos 5, and potentially GPT 5.6 in the future.
In conclusion, the speaker plans to continue using GPT-5.5 and Opus for now, given their efficiency and cost-effectiveness, while eagerly awaiting renewed access to Fable 5. Sonnet 5, despite its impressive agentic features, is seen as a more expensive and less efficient option that requires careful use. The overall sentiment is optimistic about the evolving AI landscape, with anticipation for models that combine orchestration skills with smarter decision-making to handle complex tasks more effectively.