Anthropic has unveiled its latest AI model, Claude 3.7 Sonnet, calling it its “most intelligent” to date. Released in February 2025, the large language model has “hybrid reasoning” capabilities, meaning it can think about its response — pausing to consider its answer before generating it — as well as deliver quick, real-time outputs when needed. Outperforming several of the industry’s top models, Claude 3.7 Sonnet excels at coding, software engineering, instruction-following, multimodal understanding and agentic tasks.
What Is Claude 3.7 Sonnet?
Claude 3.7 Sonnet is an AI model developed by Anthropic. With so-called “hybrid reasoning” capabilities, the model can switch between rapid responses and deep, reflective thinking within a single system.
Claude 3.7 Sonnet is hitting the market at an especially competitive moment in the artificial intelligence industry, with companies like OpenAI, Google, xAI and DeepSeek all racing to develop AI models with advanced reasoning capabilities. What sets Anthropic’s new model apart is its ability to switch between rapid responses and deep, reflective thinking within a single system. Most competing models just focus on one or the other.
“Claude 3.7 Sonnet can produce near-instant responses or extended, step-by-step thinking that is made visible to the user,” Anthropic explained in a blog post, claiming it creates a “more seamless experience” for users. “Just as humans use a single brain for both quick responses and deep reflection, we believe reasoning should be an integrated capability of frontier models rather than a separate model entirely.”
Claude 3.7 Sonnet can be accessed through the Claude chatbot across all subscription tiers (including the free tier), but its “extended thinking” mode is limited to Pro, Team and Enterprise subscribers. The model is also available through the Anthropic API, Amazon Bedrock and Google Cloud’s Vertex AI platforms.
What Is Claude 3.7 Sonnet?
Claude 3.7 Sonnet is a foundation model designed to understand and generate human-like text. Capable of providing both quick, pattern-based outputs as well as more nuanced, thought-out answers, it performs especially well in tasks involving coding, instruction-following, multimodal understanding and agentic capabilities.
Claude 3.7 Sonnet was developed by Anthropic, an AI research and development startup founded in 2021 by former OpenAI executives. Known for its Claude chatbot and large suite of language models, the company’s stated goal is to responsibly advance the field of generative AI, with an emphasis on safety and ethics. To that end, Anthropic develops some of the most advanced AI products on the market — right alongside top competitors — but does not release them until they have sufficiently met its robust safety measures.
Anthropic says it conducted extensive testing, training and evaluation of Claude 3.7 Sonnet, working with “external experts” to ensure it meets all of its security, safety and reliability standards. The company also claims Claude 3.7 Sonnet can make more nuanced distinctions between harmful and benign prompts, so it will reject or defer questions less frequently than previous Claude models.
What Can Claude 3.7 Sonnet Do?
Claude 3.7 Sonnet can do much of what other comparable models can do: answer questions, brainstorm ideas, summarize content and generate new content — accepting both images and text as inputs. But it stands out from other Anthropic models in a few important ways.
Reasoning
Claude 3.7 Sonnet is Anthropic’s first publicly available reasoning model. In general, these kinds of models are designed to break down problems into smaller, more manageable steps, verifying facts along the way before generating a final answer. While they don’t necessarily replicate human thinking or reasoning, their process is modeled after deduction, with the aim of providing more accurate and reliable responses.
Functioning as both a typical large language model and reasoning model in one, Claude 3.7 Sonnet lets users pick whether they want a quick answer from the model or whether they want it to think longer before answering.
- In “standard mode,” the model essentially functions as an upgraded version of Anthropic’s Claude 3.5 Sonnet, which excels at performing complex tasks that require rapid response, like knowledge retrieval, sales automation and computer programming.
- When “extended thinking mode” is turned on, the model creates “thinking content blocks” where it visually displays its internal reasoning to the user, according to Anthropic. Those insights are then incorporated into its final response, improving the model’s performance in math, physics, instruction-following and coding, among other tasks.
Through Anthropic’s API, users can control Claude 3.7 Sonnet’s “thinking” budget, meaning they can set a limit on how long the model should reason before responding (with a maximum of 128,000 tokens). This essentially allows them to balance speed and cost with the quality of the answer. In both standard and extended thinking modes, Claude 3.7 Sonnet costs $3 per million input tokens and $15 per million output tokens, including those used for thinking.
Coding
Anthropic says Claude 3.7 Sonnet is its best coding model to date — capable of spotting and fixing bugs, developing features, explaining technical concepts and suggesting improvements across different programming languages. Its extended thinking mode is specifically optimized for powering AI agents that can handle more complex tasks and workflows, thus accelerating the building process across the entire software development lifecycle.
In addition to Claude 3.7 Sonnet, Anthropic released a preview of its own agentic coding tool called Claude Code. Acting as an “active collaborator,” the company says the tool can search and read code, edit files, write and run tests and use command tools — all the while keeping users “in the loop.”
Anthropic claims Claude Code can complete tasks like test-driven development, debugging complex issues and large-scale refactoring — tasks that would typically take a human more than 45 minutes of manual work. In a video demonstration, the tool was able to analyze a project with a simple command like, “Explain this project structure.” Using plain English in the command line, developers were able to modify their code, with Claude Code describing its changes, testing for errors and even pushing updates to GitHub.
Claude 3.7 Sonnet Use Cases
Like all of Anthropic’s models, Claude 3.7 can be used in all sorts of ways. The company has highlighted a few in its documentation:
- Software engineering: Claude 3.7 Sonnet specifically achieves “state-of-the-art” performance on software engineering benchmarks, according to Anthropic, so it is good at solving complex software-related issues. This makes it a strong tool for tasks like code generation, debugging and automating development workflows.
- Ticket routing: The model’s advanced natural language processing capabilities can be used to automatically sort and route customer support tickets based on factors like urgency, customer intent, priority, customer profile and more.
- Customer support agent: The model’s advanced conversational abilities can be used to build automated customer support agents that can handle inquiries in real time, providing around-the-clock support and managing high request volumes with accurate responses and “positive” interactions.
- Content moderation: Specifically trained to be as “honest, helpful and harmless” as possible, the model can be used to moderate digital applications, helping to maintain a safe, respectful and productive environment.
- Legal summarization: With its advanced natural language processing capabilities, the model can efficiently summarize legal documents, extracting key information to expedite the legal research process. It can be used to review contracts, assist with litigation preparation and support regulatory work, saving users time while maintaining accuracy.
How Does Claude 3.7 Sonnet Compare to Other Models?
Anthropic compared Claude 3.7 Sonnet to other models of similar sizes and capabilities, including OpenAI’s o1 and o3-mini, DeepSeek’s R1 and xAI’s Grok 3, as well as its own Claude 3.5 Sonnet. These comparisons evaluated capabilities like software engineering, agentic tool use, instruction following, general reasoning, multimodal capabilities and agentic coding.
In short: Claude 3.7 Sonnet outperformed most of its competitors across the majority of these tests when it was in extended thinking mode. However, it scored lower than Grok 3 in graduate-level reasoning (GPQA Diamond); o1 in multilingual Q&A (MMMLU); both Grok 3 and o1 in visual reasoning (MMMU); o1, o3-mini and R1 in math problem-solving (MATH 500); and Grok 3, o1, o3-mini and R1 in high school math competition (AIME 2024). Claude 3.7 Sonnet also performed well in standard mode, but it did not beat out its competitors as consistently as when it was in extended thinking mode.
Beyond these more traditional benchmarks, Claude 3.7 Sonnet outperformed all of Anthropic’s previous models in its Pokémon gameplay tests when it was in extended thinking mode.
Claude 3.7 Sonnet Limitations
Like any other AI model, Claude 3.7 Sonnet is capable of producing inaccurate responses, and may reflect the biases present in its training data in its outputs. It also does not perform as well as other models in math-related tasks when it is in standard mode. However, it seems to experience a notable boost in this area when it is in extended thinking mode.
How to Access Claude 3.7 Sonnet
Claude 3.7 Sonnet can be accessed in several locations, including:
- Claude chatbot: Claude 3.7 Sonnet’s standard mode is available on all subscription tiers (Free, Pro, Team and Enterprise). However, its extended thinking mode is exclusive only to Pro, Team and Enterprise subscribers.
- Anthropic’s API: Users can integrate Claude 3.7 Sonnet into their own applications by accessing it through Anthropic’s API. You can learn how to use the API to build with Claude in this step-by-step guide.
- Third-party platforms: Claude 3.7 Sonnet is available on the Amazon Bedrock and Google Cloud’s Vertex AI platforms, where users can integrate and deploy the model into their own applications without having to manage the underlying infrastructure themselves.
Frequently Asked Questions
Is Claude 3.7 Sonnet available?
Yes — Claude 3.7 Sonnet is available through the Claude chatbot across all subscription tiers (including Free), but its extended thinking mode is exclusive to Pro, Team and Enterprise subscribers. Claude 3.7 Sonnet is also available through the Anthropic API, Amazon Bedrock and Google Cloud’s Vertex AI platforms.
Is Claude 3.7 Sonnet free?
Yes — a standard version of Claude 3.7 Sonnet can be accessed for free through the Claude chatbot. However, its extended thinking capabilities are only available in the Pro, Team and Enterprise subscription tiers, which vary in price. Otherwise, the model costs $3 per million input tokens and $15 per million output tokens on the Anthropic API, Amazon Bedrock and Google Cloud’s Vertex AI platforms.
Is Claude 3.7 Sonnet multimodal?
Yes — Claude 3.7 Sonnet accepts both text and image inputs, so it has multimodal capabilities. But it is only able to generate text responses.
Is Claude 3.7 Sonnet safe?
While no AI model is completely risk-free, Anthropic says it conducted extensive testing, training and evaluation of Claude 3.7 Sonnet, working with “external experts” to ensure it meets its security, safety and reliability standards. The company also claims Claude 3.7 Sonnet can make more nuanced distinctions between harmful and benign prompts, so it defers questions less frequently than prior models. Specifically, the model reduces unnecessary refusals by 45 percent in standard mode and 31 percent in extended thinking mode compared to Claude 3.5 Sonnet.
What is Claude Code?
Claude Code is an agentic coding tool developed by Anthropic that can autonomously perform advanced tasks like searching and reading code, editing files, writing and running tests, using command tools and even pushing updates to GitHub.
What is a reasoning model?
In general, reasoning models are designed to analyze complex problems, break them down into management steps and refine their responses before delivering a final answer. The goal is to provide more accurate and reliable responses than standard language models, which generate quick, pattern-based outputs. In the case of Claude 3.7 Sonnet specifically, the model can switch between rapid responses and deep, reflective thinking within a single system.
