Anthropic says its latest AI bot superior to Gemini, ChatGPT
Artificial Intelligence (AI) start-up Anthropic has unveiled its latest GenAI technology, Claude 3. The company claims its new AI model performs as well as or better than leading models such as Google's Gemini and OpenAI's GPT-4. The Claude 3 is offered in three variants—Haiku, Sonnet, and Opus—with Opus being the most powerful. According to Anthropic, these models showcase improved analysis and forecasting abilities, as well as better performance on specific benchmarks.
Claude 3 can analyze text prompts and images
Claude 3 is Anthropic's first multimodal GenAI, capable of examining both text and images. It can process photos, graphs, charts, and technical diagrams from various documents. Unlike some competitors, Claude 3 can analyze multiple images in one request (up to 20), allowing for image comparison. However, the start-up admits that the models struggle with identifying people, handling low-quality images, spatial reasoning tasks, object counting, and generating artwork.
Claude 3 will initially support 2,00,000 token context window
Anthropic says that Claude 3 can better follow multi-step instructions, deliver structured output like JSON, and converse in languages other than English compared to previous models. Initially, the model will support a 2,00,000 token context window (around 1,50,000 words). Select customers will get up to one million token context window (~7,00,000 words), making it on par with Google's newest GenAI model that offers up to a million-token context window. The larger context enables more contextually rich responses.
Improved context understanding than GPT-4
Claude 3 has demonstrated superior graduate-level reasoning than GPT-4 (50.4% vs. 35.7%) and outperformed most models in various benchmark tests, including math questions, coding, and reasoning. However, Anthropic acknowledges that Claude 3 can produce biased responses and hallucinate. Unlike certain GenAI models, Claude 3 cannot browse the web. It can only respond to inquiries using data predating August 2023. Although Claude is proficient in multiple languages, its fluency in certain "low-resource" languages may not match its proficiency in English.
Updates to improve performance
Over the next few months, Anthropic plans to roll out enhancements to the Claude 3 model family, aiming to develop a next-gen algorithm for "AI self-teaching." The company intends to add features that allow Claude to interact with other systems, code interactively, and provide advanced agentic capabilities. Anthropic is also dedicated to raising significant funding to compete with OpenAI, having already secured billions in committed capital and pledges from Google, Amazon, and other supporters.
What about Claude 3's pricing and availability?
Claude 3 Opus and Claude 3 Sonnet are now accessible on the web and through Anthropic's dev console and API, Amazon's Bedrock platform, and Google's Vertex AI. Opus cost $15 per million input tokens and $75 per million output tokens. Sonnet is priced at $3 for million input tokens against $15 for million output tokens Claude 3 Haiku will be released later this year, priced at $0.25 per million input tokens and $1.25 per million output tokens.