In this video, I look at the latest release from Anthropic of the Sonnet 4.6 model and discuss is it really the model that you want to use instead of Opus 4.6? 

Site: https://www.anthropic.com/news/claude-sonnet-4-6
ArtificialAnalysis: https://artificialanalysis.ai/evaluations/gdpval-aa

For more tutorials on using LLMs and building agents, check out my Patreon
Patreon: https://www.patreon.com/SamWitteveen
Twitter: https://x.com/Sam_Witteveen 

🕵️ Interested in building LLM Agents? Fill out the form below
Building LLM Agents Form: https://drp.li/dIMes

👨‍💻Github:
https://github.com/samwit/llm-tutorials

⏱️Time Stamps:
00:00 Intro
00:46 Sonnet 4.6 Blog
01:10 Computer Use: Claude in Chrome
01:53 Benchmarks
03:24 Pricing
03:45 Third Party Analysis
05:49 Programmatic Tool Calling

Sam Witteveen AI is a publication offering insights, tutorials, and resources for artificial intelligence (AI) enthusiasts and practitioners. Readers can learn about machine learning algorithms, deep learning frameworks, and AI applications. With tutorials, case studies, and expert interviews, Sam Witteveen AI provides  guidance and expertise for building and deploying AI solutions.

Sam Witteveen

Claude Sonnet 4.6 is Anthropic's latest model release, positioned as 40% cheaper than Opus 4.6 with improved benchmarks, computer use capabilities, adaptive thinking, and extended context up to 1M tokens. However, independent benchmarks from Artificial Analysis reveal a critical hidden cost: Sonnet 4.6 uses over 4x more tokens than its predecessor (280M vs 58M) due to adaptive thinking, and even more than Opus 4.6 (160M tokens). This 'token muncher' problem means the per-token price advantage may be entirely negated in practice. Additional concerns include uneven feature availability across API providers, with programmatic tool calling only available via Anthropic's API and Microsoft Foundry. The recommendation is to benchmark on your own workloads — Sonnet 4.6 may be cheaper for tasks not using adaptive thinking, but Opus 4.6 remains the better choice for heavy extended-thinking use cases.

The "Token Muncher" Problem: Is Sonnet 4.6 Actually Cheaper?