Anthropic/Blog/25/02/Claude 3.7 Sonnet and Claude Code

• DOING Claude 3.7 Sonnet and Claude Code \ Anthropic

  • my #notes

    • Anthropic is possibly differentiating themselves by adding something kind of like a Continuously Variable Transmission for model reasoning, rather than having to pick between a reasoning model and a traditional non-reasoning model.
      • API users can set a budget, which will be essential for controlling reasoning

        • • control LLM Reasoning Model budget through Anthropic/API - #Cool

            • through the API, users can also control the budget for thinking: you can tell Claude to think for no more than N tokens, for any value of N up to its output limit of 128K tokens. This allows you to trade off speed (and cost) for quality of answer

  • Claude 3.7 Sonnet

    • #Price

      • same as Claude 3.5 Sonnet: $3 per million input tokens and $15 per million output tokens—which includes thinking tokens

    • unified reasoning and inference

      • both an ordinary LLM and a reasoning model in one
        • pick when you want the model to answer normally and when you want it to think longer before answering.
      • standard mode
        • Claude 3.7 Sonnet is an upgraded version of Claude 3.5 Sonnet
      • Claude/Extended Thinking mode
        • self-reflects before answering
      • prompting for the model works similarly in both modes

    • control LLM Reasoning Model budget through Anthropic/API - #Cool

      • through the API, users can also control the budget for thinking: you can tell Claude to think for no more than N tokens, for any value of N up to its output limit of 128K tokens. This allows you to trade off speed (and cost) for quality of answer

    • AI Coding #Performance

      • There's a reason Claude is the default model for all Cursor users—Anthropic's approach to building models delivers on real-world tasks. During our extensive testing of Claude 3.7 Sonnet, we've seen significant improvements in the model's ability to understand and handle complex codebases and multi-step tasks. Now with two ways to think, Claude cements its place as the industry leader for coding. -- #Quote from Person/Michael Truell CursorAI Cursor, CEO
        

    • • AI Model Card
        • Claude 3.7 Sonnet excels across instruction-following, general reasoning, multimodal capabilities, and agentic coding, with extended thinking providing a notable boost in math and science. Beyond traditional benchmarks, it even outperformed all previous models in our Pokémon gameplay tests.

  • Claude Code

    • Anthropic/YouTube/25/02/Introducing Claude Code
    • limited research preview
    • AI/Coding/Tool - to delegate engineering tasks to Claude