Anthropic, the artificial intelligence company founded by former OpenAI researchers, has unveiled its latest language model called Claude 3, boasting performance that surpasses industry leaders like Google and OpenAI. In a bold claim, the company states that the new Claude 3 family of AI models can outperform OpenAI’s GPT-4 and other cutting-edge systems on various benchmarks.

The Claude 3 suite comprises three models – Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus – with Opus being the largest and most advanced offering. According to Anthropic, these models excel in a wide range of tasks, including answering complex questions, comprehending lengthy instructions, and providing accurate responses with superior contextual understanding compared to previous iterations.

One of the key advantages touted by Anthropic is Claude 3’s multimodal capabilities, allowing it to process and interpret both text and visual inputs seamlessly. This advancement positions Claude 3 as a versatile tool for tasks that require understanding and reasoning across multiple modalities.

Addressing previous limitations, Anthropic claims that the new models are less likely to refuse to answer prompts that push the boundaries of their safety guardrails. This suggests an improved ability to comprehend context and make more nuanced decisions, a trait that has been rumored to be a focus for Meta‘s upcoming Llama 3 model.

In terms of performance, Anthropic asserts that Claude 3 Opus outperformed most models, including GPT-4, in several benchmarking tests. Notably, it demonstrated superior graduate-level reasoning, achieving a 50.4 percent score compared to GPT-4’s 35.7 percent. Additionally, Claude 3 Opus excelled in math questions, coding tasks, and logical reasoning challenges.

Anthropic emphasizes the speed and efficiency of the Claude 3 models, particularly the smallest variant, Haiku. According to the company, Haiku is “the fastest and most cost-effective model on the market,” capable of rapidly processing dense research papers complete with charts and graphs in under three seconds.

The training process for Claude 3 involved a combination of nonpublic internal datasets, third-party datasets, and publicly available data as of August 2023. Anthropic utilized hardware resources from Amazon Web Services (AWS) and Google Cloud, two companies that have invested in the AI firm, with Amazon contributing a staggering $4 billion.

Claude 3 will be made available on AWS’s model library Bedrock and Google’s Vertex AI, further solidifying Anthropic’s partnerships with these tech giants. The company plans to release Claude 3 Haiku soon, while Claude 3 Sonnet and Claude 3 Opus are already accessible through claude.ai and its API.