Anthropic released an upgraded version of its Claude 3.5 Sonnet artificial intelligence (AI) model on Monday. Dubbed Claude 3.7 Sonnet, it is being made available to all Claude users. The AI firm described 3.7 Sonnet as its most intelligent model capable of advanced reasoning. The main focus of the new large language model (LLM) is coding, and to support the capability, the company also introduced Claude Code, Anthropic’s first agentic coding tool that can handle a large variety of backend coding tasks.
Anthropic Releases New AI Model and Its First AI Agent
In a newsroom post, the company announced the release of the Claude 3.7 Sonnet model. It is the first hybrid AI model by the company and can perform both as a standard language model as well as a reasoning model. Reasoning models typically utilise test-time compute functions to increase the time spent on a query. During this time, it second-guesses the output, looks for alternative solutions, and verifies the information.
With Claude 3.7 Sonnet, users can utilise the same AI model to get both standard and reasoning functions. Explaining the reason behind opting for a hybrid model, Anthropic said, “We believe reasoning should be an integrated capability of frontier models rather than a separate model entirely.”
Gadgets 360 staff members were able to access the AI model on the free tier, and the responses appear to be more sophisticated compared to the older Sonnet model. However, the improvements were marginal, which is typically the case with most iterative AI models.
Users can now access a new Thinking Mode in the model picker menu of Claude, and select between Normal and Extended. While the Normal mode will produce near-instant responses, the Extended mode will trigger reasoning-based responses. Notably, the Extended mode is currently only available to Pro subscribers.
Anthropic said developers accessing the model via the application programming interface (API) will be able to control the time the model thinks before producing an output. This can be controlled by determining a specific token value for Claude. This number can go all the way to 1,28,000 tokens, which is the upper ceiling for this model. The AI firm highlighted that this granular control will let developers build more focused products.
Coming to performance, the Claude 3.7 Sonnet scored 62.3 percent in the SWE-bench verified benchmark, outperforming the 3.5 Sonnet and OpenAI’s o1, as per the company’s internal testing. It also outperforms o1 in the TAU-bench benchmark for agentic tool use.
Additionally, the AI firm also introduced Claude Code, its first agentic coding tool in a limited research preview. It can perform a wide range of coding tasks including searching and reading code, editing files, writing and running tests, committing and pushing code to GitHub, and using command line tools.
In Anthropic’s internal testing, the agentic tool was able to complete complex tasks that more than 45 minutes of manual work in a single attempt. Interested individuals can access the preview here. The AI firm highlighted that the tool is being extensively used internally.
