Anthropic’s Claude 3 LLM challenges GPT-4 with improved “intelligence” and visual capabilities



Anthropic introduces Claude 3, its latest large language model, available in three versions. The most powerful version, “Opus”, is supposed to be at least on par with GPT-4.

AI startup Anthropic, a spin-off of OpenAI, has introduced the Claude 3 model family, a new series of AI systems designed to set standards in various cognitive task areas.

The family consists of three models: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus, offering users a choice of intelligence, speed, and cost for their specific applications. Opus is as fast as Claude 2.1, but is said to be much more “intelligent”, while Haiku can respond in near real-time. Sonnet is twice as fast as Claude 2, but with “higher levels of intelligence.”

All Claude 3 models offer improved analytics and predictive capabilities, nuanced content creation, code generation, and conversation in non-English languages such as Spanish, Japanese, and French, according to Anthropic. In addition, they can handle a variety of visual formats, including photos, charts, graphs and engineering drawings.



Opus and Sonnet are currently available through and the Claude API, with Haiku coming soon. Opus is only available to paying Claude customers, Sonnet is free.

Claude 3 models to outperform respective competitor models

According to the announcement, the Claude 3 models outperform their competitors on common AI benchmarks such as undergraduate-level expertise (MMLU), graduate-level reasoning (GPQA), and basic mathematics (GSM8K). Anthropic claims that Opus can demonstrate “near-human levels of comprehension and fluency on complex tasks.”

On the language understanding benchmark “MMLU”, Claude 3 is on par with OpenAI’s original GPT-4, which was released a year ago. | Image: Anthropic

According to Anthropic, Claude 3 models can follow complex instructions and produce structured output in formats such as JSON, making them suitable for natural language classification and sentiment analysis.

While it may be a success for Anthropic to catch up to GPT-4 in benchmarks and beat it in some, two things should be kept in mind: First, benchmarks are just that. How well the models perform in the real world remains to be seen. Second, GPT-4 has been available for about a year, and still no company has managed to make significant progress – despite all the billions invested.

Claude 3 got eyes

The new Claude models have visual capabilities that allow them to process different image formats such as photos, diagrams and technical drawings. Anthropic says this should be of particular benefit to corporate customers whose knowledge bases are encoded in various formats.


Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top