Anthropic Releases Claude 3.7 Sonnet With Extended Thinking Mode

What Happened

Anthropic released Claude 3.7 Sonnet on February 24, 2025, making it available through the company's API, the Claude.ai web interface, and third-party platforms including Amazon Bedrock and Google Cloud Vertex AI. The model is Anthropic's first to offer a hybrid architecture that allows users to toggle between a standard response mode and an extended thinking mode, in which the model works through problems step by step before delivering a final answer.

What the Model Does

In extended thinking mode, Claude 3.7 Sonnet generates an internal chain of reasoning before producing its output. Developers can set a token budget for that reasoning process, allowing them to control how long the model deliberates and, consequently, how much compute is consumed per query. Anthropic said the model demonstrates meaningful improvements on coding, mathematics, and instruction-following benchmarks compared to its predecessor, Claude 3.5 Sonnet.

On the SWE-bench Verified benchmark, which measures a model's ability to resolve real-world software engineering tasks, Claude 3.7 Sonnet scored 70.3 percent in extended thinking mode. On the AIME 2024 mathematics competition benchmark, the model scored 80 percent with extended thinking enabled. Anthropic published these figures in its model release documentation.

Pricing and Access

Anthropic priced Claude 3.7 Sonnet at three dollars per million input tokens and fifteen dollars per million output tokens, the same price as its predecessor. Tokens generated during the extended thinking process count toward output token costs. The model is available immediately to developers on paid API tiers and to Claude.ai subscribers on Pro, Team, and Enterprise plans.

Background

Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and other former OpenAI researchers. The company has positioned its Claude model family around safety research and what it calls Constitutional AI, a training methodology designed to make models more reliably aligned with human intent. Anthropic has raised several billion dollars in funding, with major investments from Google and Amazon, the latter having committed up to four billion dollars to the company.

Claude 3.7 Sonnet arrives as several competitors have released or announced reasoning-focused models. OpenAI has released its o-series reasoning models, and Google DeepMind has shipped Gemini 2.5 Pro with enhanced reasoning capabilities. The extended thinking approach Anthropic is using is comparable in concept to chain-of-thought reasoning methods used in those competing systems.

What It Means in Practice

Developers building applications that require multi-step problem solving, such as automated code review, complex data analysis, or mathematical computation, can now direct Claude 3.7 Sonnet to allocate more processing time to difficult queries while keeping faster, cheaper standard responses available for simpler tasks. The configurable token budget gives engineering teams a direct mechanism to balance response quality against cost within a single model.

Anthropic also released an updated version of its computer use capability alongside Claude 3.7 Sonnet. That feature, first introduced in October 2024, allows the model to interact with desktop graphical interfaces by interpreting screenshots and issuing keyboard and mouse commands. The updated version shows improved performance on the OSWorld benchmark, which tests an AI agent's ability to complete tasks on a standard computer desktop.

What Comes Next

Anthropic has indicated that Claude 3.7 Sonnet is part of a broader Claude 3.7 model family, with additional variants expected to follow, though the company has not announced release dates for those subsequent models.