GPT-6 Sol and Luna called underwhelming but strong on cost efficiency
The author finds Sol and Luna unimpressive but notes Luna is 60% cheaper than 5.6 Luna with better reasoning efficiency and performance, while Opus 5.5 holds the Pareto frontier and OpenAI pushes efficiency and cost.
GPT-6 Sol and Luna seem underwhelming, but I'm pretty happy with them given the usage/cost effectiveness. GPT-6 Luna is 60% cheaper than GPT-5.6 Luna while being more reasoning efficient and performing better, that is pretty nuts. GPT-6.1 will be here sooner than you think. It is sort of odd that Anthropic has the pareto frontier right now with Opus 5.5, while OpenAI is demolishing the efficiency and cost, but both have their merits, great launches by both labs! Can't wait to experiment and find the extremely niche but super effective use cases for all 3 new models.