Anthropic has made bold claims about its new Claude 3 chatbot family, with the AI startup insisting it can outperform OpenAI’s ChatGPT. The US-based firm unveiled three new benchmark-setting models earlier this week, including Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus.
The new models differ in levels of speed and power, with ‘Haiku’ the smallest in terms of compute capability and ‘Opus’ the largest.
Anthropic revealed that its ‘Opus’ model “outperforms its peers on most of the common evaluation benchmarks,” including undergraduate-level expert knowledge (MMLU) and graduate-level expert reasoning (GPQA).
Among the rival models it bested on these parameters were the likes of OpenAI’s GPT-4 and GPT-3.5, as well as Gemini 1.0 Ultra and Gemini 1.0 Pro.
All the new Claude 3 models also show “increased capabilities” in areas of nuanced content generation, code generation, and non-English languages like Spanish, Japanese, and French.
While OpenAI’s chatbot offerings are frequently the ones grabbing the headlines, these new models reinforce the fact the Microsoft-backed startup now faces increasingly stiff competition, according to Arun Chandrasekaran, VP analyst at Gartner.
Speaking to ITPro, Chandrasekaran said the latest announcement shows that 2024 could be a whirlwind year for generative AI development as competition between major industry players heats up.
“This is promising to be another crazy year in terms of just [the] amazing pace of innovation, and [the amount of] new developments in this space,” he said.
A notable aspect of this announcement, Chandrasekaran said, was the enhanced context window Anthropic was keen to highlight. Claude 3 will boast a “200K context window upon launch”, the firm said.
Anthropic also said all three new models are capable of accepting inputs exceeding 1 million tokens, a feature they may make accessible to select customers who need “enhanced processing power.”
Anthropic is bullish on “constitutional AI”
Chandrasekaran said an exciting aspect of this announcement is Anthropic’s well-publicized commitment to “constitutional AI”.
The firm has been keen to frame itself as a more responsible, ethical-led player in the generative AI space than some industry counterparts in recent months.
“One of the things that Anthropic has consistently spoken about since its founding is this notion of constitutional AI,” Chandrasekaran said.
“This whole view that they want to align the AI models very closely with human intent and human alignment,” he added. “I think that’s something that makes Anthropic a little different”
Google and AWS have also pledged major support for Anthropic in recent months, investing over $5 billion in the firm in a clear sign that both hyperscalers view it as a leading contender to OpenAI.
Anthropic isn’t alone in gaining ground on OpenAI
It’s not just Anthropic announcing new models and putting pressure on the reigning champion OpenAI.
Google, despite playing catch-up to Microsoft and OpenAI over the last year, appears to be gaining ground on the duo’s dominance in the generative AI space.
It recently brought Duet AI chatbot under the united roof of Gemini, which now looks poised to make a considerable impact on the AI landscape through multimodality.
“Google, with the Gemini 1.5 model in particular, announced some very impressive capabilities,” Chandrasekaran said.
“The most impressive [sic] is the million context window for the model,” he added.
This, Chandrasekaran explained, means that users can work with huge amounts of text, speech, and video on the platform. He also mentioned Meta, and the impressive capabilities of its own open source model offerings.
Analysts told ITPro last month that the Gemini rebranding marked a significant moment in the generative AI boom, and one that will lay the groundwork for a looming race between Microsoft and Google.
Chirag Dekate, VP analyst at Gartner, said the new model range from the tech giant could supercharge Google’s attempts to catch Microsoft in the battle for AI dominance.
