Anthropic and OpenAI launched new, lower-cost artificial intelligence models on Tuesday, just hours apart, days after executives from both companies urged caution in developing the technology.
Anthropic stated that its new Opus 5.5 model approaches the performance of its flagship model at a significantly lower cost, explaining that it matches the advanced Vable model in most tasks and at a 20% discount.
Subsequently, OpenAI announced the launch of its GPT-6 Sol and GPT-6 Luna models, asserting that they perform tasks far better than Anthropic's most powerful models.
The new models are faster and lower-cost versions of the GPT-6 Astra, OpenAI's most powerful model, which was released earlier this month.
This competition reflects the dilemma facing both companies: they warn of the risks associated with the technology they are developing, while simultaneously under pressure to generate returns on their massive investments amidst fierce competition, particularly from low-cost Chinese models.
Anthropic launched its new model weeks before its anticipated IPO, while OpenAI postponed its initial public offering plans until next year.
Anthropic confirmed that the new model employs the same safeguards used in Fable across cybersecurity, biology, and AI model design.
It added that internal testing showed Opus 5.5 to be the best model the company has tested to date in terms of "conformity"—its adherence to the goals set by its developers.
However, the company also noted that the model exhibited a recurring tendency to believe it was being evaluated, which it acknowledged could make its behavior under normal usage conditions difficult to predict.
0 Comments