A report by AI investment outfit Air Street Capital said that OpenAI still retains an edge on reasoning tasks with the release of its o1 "Strawberry" model—which Air Street's report rightly characterised as a weird mix of incredibly strong logical abilities for some tasks and surprisingly weak ones for others.

The report’s author, Nathan Benaich, said the cost of using a trained AI model is falling rapidly because models are differentiated based on capabilities and performance and must compete on price.

Another reason is that engineers for companies such as OpenAI and Anthropic -- and their hyperscale partners Microsoft and AWS-- are discovering ways to optimise how the most significant models run on significant GPU clusters. The cost of outputs from OpenAI's GPT-4o today is 100 times less per token (equivalent to 1.5 words) than GPT-4 when that model debuted in March 2023. Google's Gemini 1.5 Pro now costs 76 per cent less per output token than when that model was launched in February 2024.

AI researchers have also become good at creating small AI models that can equal the performance of larger LLMs on dialogue, summarisation, or even coding while being much cheaper to run.

These two trends mean that the economics of implementing AI-based solutions are starting to look much more attractive than a year ago. This may ultimately help businesses find the return on investment from generative AI, which they have complained has been elusive.