Anthropic to Soon Release Claude 3.5 Opus

Anthropic is planning to release Claude 3.5 Opus soon, revealed its chief Dario Amodei in the latest episode of Lex Fridman podcast. 

This development comes in the backdrop of Anthropic releasing Claude 3.5 Sonnet and Claude 3.5 Haiku in July. However, Anthropic did not release an iterative upgrade for its most capable model, Opus. This raised questions in the AI community as to whether Anthropic felt that Opus was redundant. 

“Not giving you an exact date, but as far as we know, the plan is still to have a Claude 3.5 Opus,” said Amodei in response to Fridman’s question about the release of Claude 3.5 Opus. 

Anthropic’s plan seems to first ‘shift’ the capabilities of existing models. Amodei said that the Sonnet was the ‘middle child’ among all the models, but Sonnet 3.5 was much more capable than Opus 3 and cost the same as the Sonnet 3 model. He even went on to say that Haiku 3.5, its smallest model, is as powerful as Opus 3.

“So basically the aim here is to shift the curve and then at some point there’s going to be an Opus 3.5,” said Amodei. 

In the podcast, he explains that the delay is due to the need for excess computational power in pre-training, rigorous safety testing, and Anthropic’s desire to build a new version of Opus that solves newer problems to deliver an unprecedented value. 

“That (pre-training) uses, these days, tens of thousands, sometimes many tens of thousands of GPUs or TPUs or training them, or we use different platforms, but accelerator chips, often training for months,” said Amodei. 

So Anthropic does see an opportunity to further scale their models, with their most powerful model set to release soon. So is it still fair to conclude LLMs are hitting the wall? 

That said, Anthropic has been on a feature release spree over the past few weeks. The new Computer Use feature took the AI community by storm, which may well change the state of human-computer interaction. They also launched a Visual PDF tool that lets Claude visually analyse PDF documents, and provide insights regarding the images, charts, and graphs inside. 

With OpenAI set to release o1 in the next few days, and Google heading towards the release of Gemini 2.0 at the end of this year, we’ll need to buckle up for a new battle of the flagship models.