AI startup Anthropic has unveiled the latest version of its GenAI technology, named Claude 3, claiming that it rivals the performance of OpenAI’s GPT-4.
Backed by substantial venture capital, Anthropic’s new GenAI comes in three variants: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus, with Opus being the most powerful in the lineup. Anthropic asserts that these models exhibit increased capabilities in analysis and forecasting compared to GPT-4 and Google’s Gemini 1.0 Ultra.
What sets Claude 3 apart is its status as Anthropic’s first multimodal GenAI, capable of analyzing both text and images. It can process a variety of visual data, including photos, charts, graphs, and technical diagrams sourced from PDFs, slideshows, and other document types. Importantly, Claude 3 introduces the ability to analyze multiple images in a single request, allowing for image comparison.
Claude 3 vs OpenAi
However, Anthropic has set limits on Claude 3’s image processing capabilities. The models are restricted from identifying people, addressing ethical and legal concerns. Additionally, Claude may encounter challenges with low-quality images (under 200 pixels) and tasks involving spatial reasoning or object counting.
Anthropic emphasizes that Claude is designed to excel in following multi-step instructions, producing structured output, and engaging in conversations in languages other than English. The company claims improvements in response expressiveness and engagement, making it more effective with concise prompts.
Claude’s expanded context is highlighted as a significant enhancement. The model initially supports a 200,000-token context window, equivalent to about 150,000 words, with select customers receiving up to a 1-million-token context window (~700,000 words). This move aligns with Google’s Gemini 1.5 Pro, offering a similar context window size.
Despite its advancements, it acknowledges some challenges common to GenAI models, including bias and hallucinations. Unlike some competitors, Claude 3 cannot search the web and relies on data available only until August 2023. It also has varying proficiency in certain “low-resource” languages compared to English.
Anthropic promises regular updates to Claude 3 in the coming months, expressing confidence in the model’s ongoing evolution. The company envisions building an algorithm for “AI self-teaching,” aiming to create virtual assistants capable of handling various tasks, similar to existing large language models.
As for pricing, Anthropic offers a breakdown for the different variants:
- Opus: $15 per million input tokens, $75 per million output tokens
- Sonnet: $3 per million input tokens, $15 per million output tokens
- Haiku: $0.25 per million input tokens, $1.25 per million output tokens
Anthropic’s ambitious goals, commitment to regular updates, and its unique approach to training GenAI with “constitutional AI” make it a notable player in the competitive AI landscape. The company aims to raise significant capital, signaling its determination to stay competitive with industry giants like OpenAI. As Anthropic continues to evolve, the AI community awaits further developments and innovations in the GenAI space.