Coming soon

Market insights

Coming soon




Anthropic claims its new AI chatbot models beat OpenAI’s GPT-4

4 mins

Daniil Bazylenko

Published by: Daniil Bazylenko

18 March 2024, 10:40AM

In Brief

Anthropic, backed by Google and significant venture capital, has unveiled Claude 3, an AI chatbot claiming superiority over OpenAI's GPT-4.

Claude 3 is Anthropic's first multimodal GenAI, analyzing both text and images, with models like Haiku, Sonnet, and Opus, the most powerful among them.

The models show increased capabilities, outperforming benchmarks set by ChatGPT, GPT-4, and Google's Gemini 1.0 Ultra (excluding Gemini 1.5 Pro).

Despite advancements, Claude 3 has limitations, such as not identifying people due to ethical concerns and challenges with low-quality images.

Anthropic plans frequent updates, aiming to enhance model intelligence, and Claude 3 models are available, with Haiku expected later, alongside ambitious fundraising goals of up to $5 billion.

Anthropic claims its new AI chatbot models beat OpenAI’s GPT-4

Anthropic, a startup supported by Google and backed by substantial venture capital, has unveiled its latest AI chatbot, Claude 3, asserting its superiority over OpenAI's GPT-4. Claude 3, Anthropic's first multimodal GenAI, introduces models like Haiku, Sonnet, and Opus, with Opus being the most potent. These models exhibit "increased capabilities" in analysis and forecasting, outperforming benchmarks set by ChatGPT, GPT-4, and Google's Gemini 1.0 Ultra (excluding Gemini 1.5 Pro). A significant feature is Claude 3's ability to analyze both text and images, with improved image processing, allowing the simultaneous analysis of up to 20 images in a single request.

While Claude 3 demonstrates advancements in processing images, Anthropic has imposed limits, preventing the models from identifying people due to ethical and legal concerns. Additionally, the chatbot may encounter challenges with low-quality images (under 200 pixels), spatial reasoning tasks, and object counting in images. The models strictly focus on image analysis, refraining from generating artwork, at least in the current iteration.

Anthropic assures that customers can anticipate Claude 3's enhanced performance in following multi-step instructions, producing structured output in formats like JSON, and supporting conversations in languages beyond English compared to its predecessors. Claude 3 also aims to decrease the instances of refusing to answer questions, thanks to its "more nuanced understanding of requests." Future updates will include features to allow the models to cite the source of their answers, providing users with a means of verification.

The improvements in Claude 3 are attributed, in part, to its expanded context window. Anthropic states that Claude 3 will initially support a 200,000-token context window, equivalent to approximately 150,000 words. Some select customers will have access to a 1-million-token context window (~700,000 words), placing it in line with Google's Gemini 1.5 Pro.

While Claude 3 represents a notable upgrade, Anthropic acknowledges the persisting challenges faced by GenAI models, such as bias and hallucinations. Unlike some competitors, Claude 3 lacks the ability to search the web and can only answer questions using data available up to August 2023. Additionally, its proficiency in certain "low-resource" languages may not match its fluency in English.

Despite these limitations, Anthropic is committed to regular updates for Claude 3 in the coming months. The company believes that there is considerable room for improvement in model intelligence and plans to enhance the Claude 3 model family over time.

Claude 3 models, namely Opus and Sonnet, are currently available on the web and via Anthropic's development console and API, Amazon's Bedrock platform, and Google's Vertex AI. Haiku is expected to be introduced later this year.

The pricing structure for Claude 3 includes Opus at $15 per million input tokens and $75 per million output tokens, Sonnet at $3 per million input tokens and $15 per million output tokens, and Haiku at $0.25 per million input tokens and $1.25 per million output tokens.

Looking at the broader context, Anthropic's ultimate ambition is to develop a next-gen algorithm for "AI self-teaching." This algorithm could potentially power virtual assistants capable of handling tasks such as answering emails, conducting research, and generating various forms of content. While drawing parallels with competitors like GPT-4, Anthropic is keen on incorporating features that enable Claude 3 to interact with other systems, code "interactively," and deliver "advanced agentic capabilities."

Anthropic is in it for the long haul, having set ambitious fundraising targets of up to $5 billion over the next 12 months, as indicated in a leaked pitch deck from May of the previous year. With commitments and pledges from major players like Google and Amazon, Anthropic is well on its way to achieving these financial milestones, securing its position in a competitive landscape dominated by the likes of OpenAI.

User Comments

There are no reviews here yet. Be the first to leave review.

Hi, there!

Join our newsletter

Stay in the know on the latest alpha, news and product updates.