Anthropic presents Claude 3.7 Sonnet its first hybrid reasoning model with improvements in coding and mathematics

Anthropic He has presented his new artificial intelligence model (AI) Claude 3.7 Sonnet who defines as “first model of hybrid reasoning “, since it allows to obtain “almost snapshot” responses or results elaborated with an extended thought, even, with the possibility of controlling the time during which the model may think.

The company has released its Model of smarter to date that, in addition to presenting “important” improvements in sectors such as coding and web development ‘Front-End’, it has been developed with a “different philosophy” to other market reasoning models.

In this case, Claude 3.7 Sonnet is both a model of Large language (LLM) Like a Reasoning model “all in one”, which offers a unified approach that creates a more fluid experience for users.

That is, as Anthropic explained in a statement on their website, as well as humans use “a single brain” to develop rapid responses and deep reflections, the Reasoning “should be an integrated capacity of avant -garde models”, instead of presenting itself as a separate model.

Therefore, Claude 3.7 Sonnet allows users They choose when they want the model to respond quickly and when they want you to think more before answeringthrough a drop -down menu that allows you to select the “mode of thought” for each consultation.

Likewise, the company has also pointed out that users who have access to the API can Control the “budget to think”. Specifically, they may indicate to Claude that they think for a certain number of tokens, until its output limit of 128 tokens. In this way, they can balance speed and cost, optimizing the quality of the response.

In standard mode, the new model presents an improved version of Claude 3.5 Sonnet, with Improvements in its coding skills and web development ‘Front-End’among other novelties. For its part, in the mode of Extended thoughtClaude 3.7 Sonnet is able to reflect before responding, improving his performance in consultations of Mathematics, Physics and Instruction Monitoringamong other tasks.

Likewise, the technological one has indicated that, when developing Claude 3.7 Sonnet, they have focused especially on the development of “real world” tasks, that reflect how companies really use the LLMs.

As for the tests carried out by Anthropic, the new model has stood out in the monitoring of instructions, the general reasoning, multimodal capacities and agent coding, in addition to offering a remarkable impulse in mathematics and coding when using extended thinking.

With all this, Anthropic has pointed out that Claude 3.7 Sonnet is now available for all users signed to Claude’s plans, including the free, PRO, Team and Enterprise. Likewise, the new model can also be accessed through the Anthropic API, Amazon Bedrock y Vertex Ai by Google Cloud.

Likewise, the technological one has clarified that both the standard mode of thought of Claude 3.7 Sonnet, as the extended one, have the same price as your predecessors. That is, 3 dollars (around 2.87 euros to change) for one million entrance tokens and 15 dollars (14.34 euros) for every million departure tokens.

CLAUDE CODE

In addition to all this, Anthropic has also presented a new online command tool for agents codingto which it refers as Claude Codeand that, for the moment, it is in a sight of limited investigation.

This tool is based on a Active collaborator who can search and read code, as well as edit files, write and execute tests. Likewise, the technology has underlined its ability to confirm and send code to GitHub and use command line tools, informing the user at each step.

Although it is an initial product, Anthropic has highlighted its use for testing based on tests, the purification of complex problems and the “large -scale refactorization”. Thus, in the first tests, Claude Code He completed tasks that would normally carry more than 45 minutes of manual work.

Thus, the technological one has detailed that this agents coding tool will continue to improve to add support for prolonged execution commands and improve representation in the application, among other issues.