Nowadays, the field of large models is even bigger! OpenAI released GPT4o on the front foot and controlled the whole game, but was surpassed by the biggest competitor Anthropic on the back foot. 
Just now, Anthropic released a new large model Claude 3.5 Sonnet, which is claimed to be the smartest model so far. 
According to reports, Claude 3.5 Sonnet is the first version of Anthropic’s upcoming Claude 3.5 series. This model has improved the level of intelligence in the entire field, surpassing competing large models and its most powerful predecessor, the Claude 3 Opus, in most benchmark evaluations. At the same time, the running speed and cost are equivalent to that of its predecessor Claude 3 Sonnet. Address: https://claude.ai/Currently, Claude 3.5 Sonnet is already available for free on Claude.ai and the Claude iOS app, while Claude Pro and Team plan subscribers can get it for a higher price rate to access it. In addition, Claude 3.5 Sonnet can already be used directly through Anthropic API, Amazon Cloud Technology Bedrock and other channels. The price is $3 per million input tokens, $15 per million output tokens, and the context window is 200k tokens. 
Anthropic says it aims to dramatically improve the trade-off curve between performance, speed and cost every few months. The Claude 3.5 Haiku and Claude 3.5 Opus will be released later this year. In addition to developing the next generation of large models, Anthropic is developing new patterns and capabilities to support more enterprise use cases, including integration with enterprise applications. The team is also exploring features such as Memory, which allows the model to remember the user's preferences and specified interaction history, enabling a more personalized and efficient experience. Twice faster than Claude 3 Opus Claude 3.5 Sonnet For Graduate Level Reasoning (GPQA), Undergraduate Level Knowledge (MMLU) and Coding Capabilities (HumanEval) sets new industry benchmarks . It shows significant improvements in understanding nuance, humor, and complex instructions, and excels at writing high-quality content in a natural, approachable tone. 
Claude 3.5 Sonnet runs twice as fast as Claude 3 Opus. This increase in performance combined with an affordable price makes the Claude 3.5 Sonnet ideal for complex tasks. In an internal proxy coding evaluation, Claude 3.5 Sonnet solved 64% of the problems, better than Claude 3 Opus (38% solved). The Anthropic evaluation tested Claude 3.5 Sonnet's ability to fix bugs or add functionality to the open source code base based on natural language descriptions of requirements. After receiving guidance and relevant tools, Claude 3.5 Sonnet can independently write, edit and execute code with sophisticated reasoning and troubleshooting capabilities. Claude 3.5 Sonnet can easily handle code conversion, which is especially effective for updating legacy applications and migrating code bases. The following table compares the Claude 3.5 Sonnet with other models. The evaluation benchmarks include reasoning, reading comprehension, mathematics, science and coding. In all these benchmarks, the Claude 3.5 Sonnet outperforms the previous leading-edge model, the Claude 3 Opus. We can also see the comparison data between Claude3.5 and GPT-4o in the table. Except for mathematics, the scores are higher or close to than GPT-4o. 
The strongest visual model to dateAnthropic says the Claude 3.5 Sonnet is its strongest visual model, surpassing its predecessor Claude 3 Opus in all standard visual benchmarks. Performance is especially evident in tasks that require visual reasoning, such as interpreting charts, graphs, or transcribing text from imperfect images, a core function in the retail, logistics, and financial services industries. In these areas, AI can glean more insights from images, graphics, or illustrations than from text alone. The picture below shows the performance comparison between Claude 3.5 Sonnet and Claude 3 Opus, GPT-4o, Gemini 1.5 Pro in visual mathematical reasoning, scientific charts, visual question answering, chart QA, and document visual QA tasks. 
Artifacts – New ways to use Claude Anthropic also launched Artifacts on Claude.ai, a new feature that expands how users interact with Claude. When users ask Claude to generate something like a code snippet, text document, or website design, these artifacts appear in a dedicated window next to their conversation. This creates a dynamic workspace where they can view, edit and build on Claude’s creations in real time, seamlessly integrating AI-generated content into their projects and workflows. The Artifacts preview feature marks Claude’s evolution from conversational AI to a collaborative work environment, and will soon be expanded to support team collaboration. In the near future, teams will be able to securely centralize their knowledge, documentation, and work-in-progress in a shared space, with Claude acting as a teammate on demand. Finally, Anthropic says they rigorously test the model and work to reduce its potential for abuse. The Anthropic security team conducted a series of assessments on Claude 3.5 Sonnet in the areas of chemical, biological, radiological and nuclear (CBRN) risks, cybersecurity and autonomous capabilities, and finally classified Claude 3.5 Sonnet as AI Security Level 2 ( ASL-2) model, indicating that it does not pose a risk of catastrophic injury. Additionally, as part of its commitment to security and transparency, Anthropic is working with external experts to test and refine the security mechanisms in this latest model. Official blog address: https://www.anthropic.com/news/claude-3-5-sonnetThe above is the detailed content of Beat GPT-4o! The new king Claude 3.5 Sonnet is here, available directly for free. For more information, please follow other related articles on the PHP Chinese website!
Statement：The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn
