Anthropic has announced the launch of the Claude 3.5 Sonnet model, which works at twice the speed of Claude 3 Opus, can transcribe texts with more natural language and has improved when it comes to understanding nuances, humor and managing complex instructions.
The company presented this family of AI models last March, when it announced that it consisted of Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus – in ascending order according to their capacity, which offer increasingly powerful performance to adjust to the needs of users and achieve an optimal balance between intelligence, speed and price.
Then, he noted that all of them were capable of maintaining conversations, autocompleting text, carrying out data extraction tasks and processing a wide range of visual formats, such as photographs, tables or technical diagrams.
Anthropic has now announced a new iteration of the Sonnet model, the latest version of which is the Claude 3.5 and which above by Claude 3 Opus in different evaluations and speed, as well as “to the competition models”, according to the technological.
Claude 3.5 Sonnet is now available for free on Claude.ai and the Claude iOS device app, while subscribers to the Claude Pro and team plans will be able to access it with “significantly higher” speed limits, according to Anthropic.
This model sets new benchmarks for Graduate Level Expert Reasoning (GPQA), Undergraduate Level Expert Knowledge (MMLU), and the HumanEval coding test. Likewise, you improve your understanding of nuances, humor, and complex introductions, according to Anthropic.
This has also indicated that it improves in writing high-quality content with a natural and easy-to-understand tone, in addition to running at twice the speed of Claude 3 Opus and presenting itself as the “ideal” option for complex tasks, such as customer service and organizing multi-stage workflows.
Likewise, the firm has indicated that, when given the relevant instructions and tools, this model can write, edit and execute code independently with Sophisticated reasoning and problem-solving abilities.
To highlight these data, the firm has carried out internal evaluations, which have shown that Claude 3.5 Sonnet solved 64 percent of the problems, surpassing Claude 3 Opus, which managed to decipher 38 percent.
On the other hand, Anthropic has said that Claude 3.5 Sonnet can accurately transcribe text from imperfect images and now obtain more information from an image, graph or illustration.
The model also integrates with the Artifacts feature through Claude.ai, a feature that expands the possibilities of interaction with Claude and which now appear in a dedicated window next to the text box when the user requests to generate content, such as code snippets or documents. Likewise, you can request website design.
The model is also available through the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI. Its price is 3 dollars (2.80 euros at the exchange rate) per million input tokens and 15 dollars (about 14 euros) per million output tokens, with a context window of 200,000 tokens.
Finally, the company has stressed that it aims to “substantially improve the balance curve between intelligence, speed and cost every few months” and that it will launch the Claude 3.5 Haiku and Claude 3.5 Opus models later this year.