Anthropic Announces Claude 3 LLM Family

Anthropic today unveiled its Claude 3 family of large language models (LLMs), offering improved performance and a range of capabilities.

“The Claude 3 model family sets new industry benchmarks across a wide range of cognitive tasks,” Anthropic’s announcement post explains. “The family includes three state-of-the-art models in ascending order of capability: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. Each successive model offers increasingly powerful performance, allowing users to select the optimal balance of intelligence, speed, and cost for their specific application.”

Claude 3 Opus and Sonnet are available now, the firm notes, while Claude 3 Haiku will be available soon. Opus is Anthropic’s most intelligent language model, and it outperforms OpenAI GPT-4 and Google Gemini 1.0 Ultra in numerous AI benchmarks, including undergraduate-level expert knowledge (MMLU), graduate-level expert reasoning (GPQA), basic mathematics (GSM8K), and others. Anthropic adds that Opus “exhibits near-human levels of comprehension and fluency on complex tasks, leading the frontier of general intelligence.”

The performance of Claude 3 Opus is similar to that of Claude 2 and Claude 2.1, but with much higher levels of intelligence, Anthropic claims. Meanwhile, Claude 3 Sonnet is about twice as fast as Claude 2 and Claude 2.1 in most workloads while offering higher levels of intelligence, and it’s tuned for knowledge retrieval, sales automation, and other tasks that demand rapid responses.

Claude 3 Haiku will be the fastest and most cost-effective LLM “on the market” when it becomes available, Anthropic says. “It can read an information and data-dense research paper on arXiv (~10k tokens) with charts and graphs in less than three seconds.”

All three of the Claude 3 models also offer sophisticated vision capabilities that Anthropic says are on par with the visual capabilities of other leading models like OpenAI ChatGPT-4V and Google Gemini 1.0 Ultra. The models can process photos, charts, graphs, technical diagrams, and other visual formats, and Anthropic expects this capability to prove popular with its enterprise customers, some of which have up to 50 percent of their knowledge bases PDFs, flowcharts, presentation slides, and other non-textual formats.

The new models are also more accurate, trustworthy, and reliable. “The Claude 3 models show a more nuanced understanding of requests, recognize real harm, and refuse to answer harmless prompts much less often” the firm notes. The models are also easier to use, and better at handling complex, multistep instructions.

Claude 3 Haiku, Sonnet, and Opus are each launching with 200K context windows, but all three can handle inputs exceeding one million tokens, so Anthropic is looking into how it can expand the context window for customers that need enhanced processing power.

You can learn more at the Anthropic Claude website.