Anthropic releases next generation of Claude, claims it is ‘superior’ to GPT-4 Anthropic releases next generation of Claude, claims it is ‘superior’ to GPT-4

Anthropic releases next generation of Claude, claims it is ‘superior’ to GPT-4

Anthropic said the new suite of its GenAI models beat the benchmark results produced by GPT-4 and Gemini 1.0 Ultra.

Anthropic releases next generation of Claude, claims it is ‘superior’ to GPT-4

Cover art/illustration via CryptoSlate. Image includes combined content which may include AI-generated content.

Anthropic is launching its latest generative AI (GenAI) technology, Claude 3, with the claim that it is superior to OpenAI’s Chat GPT-4, according to a March 4 press release.

The new suite of models, comprising Claude 3 Haiku, Claude 3 Sonnet, and the most advanced, Claude 3 Opus, is positioned to rival OpenAI’s GPT-4 in performance metrics.

Superior performance

Supported by significant venture capital investments, Anthropic’s announcement highlighted Claude 3’s enhanced analytical and forecasting abilities, claiming superior performance on certain benchmarks compared to GPT-4 (excluding GPT-4 Turbo) and Google’s Gemini 1.0 Ultra (excluding Gemini 1.5 Pro).

A noteworthy advancement is Claude 3’s multimodal functionality, which enables it to process and analyze text and images, aligning it with capabilities found in some GPT-4 and Gemini versions. This includes processing up to 20 images in a single inquiry, a feature aimed at facilitating comprehensive image analysis.

However, Anthropic has restricted Claude 3’s image processing to mitigate potential ethical and legal concerns, explicitly avoiding individual identification. The company has acknowledged limitations in the model’s performance with low-quality images and tasks requiring spatial reasoning or precise object counting.

Claude 3’s primary function is image analysis rather than artwork creation. The model reportedly improves upon previous versions’ ability to follow multi-step instructions, generate structured outputs in formats like JSON, and offer multilingual support with a refined understanding of user requests.

With initial support for a 200,000-token context window, Claude 3 aims to deliver more contextually rich responses, comparable to Google’s latest Gemini model, which also supports a significant token context window.


Anthropic has candidly addressed the limitations of Claude 3, noting its susceptibility to biases and hallucinations — issues common in current GenAI models. The absence of web search capabilities restricts Claude 3 from using data available before August 2023. While the model offers multilingual support, it faces challenges with specific languages due to limited resources.

The company plans to regularly update Claude 3 to enhance its capabilities and address its limitations. Opus and Sonnet models are currently available for use, with the Haiku model expected to be released later in the year. Anthropic has also provided a pricing structure for these models, aiming to cater to a diverse user base.

In the broader context, Anthropic aims to lead in developing AI self-teaching algorithms, potentially transforming the capabilities of virtual assistants and other AI applications. The company is also exploring “constitutional AI” to make AI behavior more predictable and aligned with human intentions.

Mentioned in this article
Posted In: AI, Featured, Technology