Anthropic updates Claude 3.5 Sonnet with coding improvements and releases Claude 3.5 Haiku, along with an innovative beta for desktop use.
Anthropic has launched important updates and new developments in its Claude line of artificial intelligence models. The company introduced an improved version of Claude 3.5 Sonnet, which offers significant advances compared to its predecessor.
In addition, Anthropic announced a new model called Claude 3.5 Haiku, which stands out for its speed and affordable cost, along with the introduction in public beta of a revolutionary “computer use” capability, which allows developers to direct Claude to Interact with computers as a human would.
Claude 3.5 Sonnet: Coding Power
The improved version of Claude 3.5 Sonnet shows progress in all areas, but stands out especially in coding tasks. Based on testing by Anthropic, this model outperforms all publicly available systems in specific coding and tool usage benchmarks.
In the SWE-bench Verified, for example, its performance increased from 33.4% to 49.0%, even surpassing systems specialized in advanced coding such as OpenAI o1-preview.
The company has collaborated with clients such as GitLab to test the new model in DevSecOps tasks, reporting a 10% increase in reasoning skills without increasing latency. Other customers, such as Cognition, have seen notable improvements in coding, planning, and problem solving, surpassing results from previous versions.
Claude 3.5 Haiku: Speed and Efficiency
The new Claude 3.5 Haiku model, launching this month, is an update to the faster model of the previous generation. Compared to Claude 3 Opus, Haiku 3.5 matches its performance in multiple evaluations, but with a similar cost and speed to its predecessor, Claude 3 Haiku.
It has excelled in coding tasks, scoring 40.6% in the SWE-bench Verified, outperforming agents using state-of-the-art public models.
Anthropic plans to make this model available in its API, on Amazon Bedrock, and on Google Cloud’s Vertex AI. It will initially be text-only, but is expected to expand to image inputs in the near future.
Computer Use: Innovation in Public Beta
The most revolutionary functionality that Anthropic has released in public beta is the ability to use computers. Claude 3.5 Sonnet is the first model that allows developers to direct their actions to interact with a computer, using standard user interfaces and software in the same way that a person would. This feature allows you to automate repetitive processes, test software, and perform open tasks such as research.
Anthropic mentions that the beta is still experimental and has certain limitations, such as difficulty performing seemingly simple actions such as scrolling, dragging or zooming. However, companies such as Asana, Canva and Replit have already begun testing this feature to optimize their processes and products, indicating that it could have a significant impact on the efficiency and autonomy of artificial intelligence.
Safety and Ethics in the Use of AI
To ensure safe use of these new capabilities, Anthropic has conducted joint testing with the US AI Safety Institute (US AISI) and its counterpart in the United Kingdom (UK AISI). The company continues to implement its responsible scaling policies and has evaluated the potential risks of this model.
In addition, they have developed classifiers that identify the use of computer usability and detect any malicious activity.
Looking to the Future
Anthropic is enthusiastic about the possibilities these new tools and capabilities will open up. The company has invited developers to experiment with the public beta and provide feedback to improve and evolve the technology. The long-term goal is to better understand the potential and implications of increasingly advanced AI systems.
The new version of Claude 3.5 Sonnet is now available to all users through the Anthropic API, Amazon Bedrock and Google Cloud’s Vertex AI. For its part, Claude 3.5 Haiku will be released in the coming weeks, continuing the expansion and improvement of Anthropic’s line of artificial intelligence products.
Anthropic continues as one of the leaders in the development of advanced artificial intelligence with its Claude 3.5 Sonnet and Haiku models. Improvements in coding and innovation in the use of computers represent a significant step towards the automation of complex tasks, offering developers new tools to explore the potential of these technologies.
Source: geeksroom.com
