OpenAI Unveils Most Efficient Token Yet: GPT-5.6 Sol Uses Half the Tokens to Match Fable 5

According to Watchful AI monitoring, Artificial Analysis has released the latest Token Efficiency Comparison. This month, more than 5 labs have intensively launched new models, with OpenAI still holding the majority of the Pareto frontier.

The Pareto frontier means that under equal intelligence, no other model can achieve the same level of efficiency with fewer Tokens. GPT-5.6 Sol scored 59 points, averaging about 15,000 output Tokens per task. Claude Fable 5 scored 60 points, consuming around 33,000 Tokens per task.

The unit price of GPT-5.6 Sol's output is $30 per million Tokens, which is not cheap. However, in this test, the cost per task is still only $1.04, about a third of Fable 5's cost. OpenAI's in-house models Terra and Luna did not enter this efficiency frontier.