GPT-6 Sol and Luna Landed. Here's What DevDay Brings Next

On September 22, OpenAI unveiled two new models, GPT-6 Sol and GPT-6 Luna, alongside a significant price reduction of their API. A week later, DevDay 2026 commenced in San Francisco, unveiling a dozen or so fresh product launches. Sam Altman hinted at a new development during the conference, stating, "We have found a new thing." However, the focus of DevDay was not merely on the new models, but rather on the infrastructure that enables them to operate efficiently at scale.

GPT-6 Sol and Luna are not merely cost-optimized siblings of OpenAI's frontier model, GPT-6 Astra. They inherit Astra's advancements in reasoning, coding, and computer use, and are tailored for production workloads. GPT-6 Sol serves as the everyday workhorse, capturing nearly 95% of Astra's exam accuracy at about 20% of the cost, while GPT-6 Luna acts as the volume play, matching GPT-5.6 Sol accuracy at one-hundredth the cost. Both models are available across ChatGPT Work, Codex, and the OpenAI API.

Despite the aggressive 90% input discount, the real value of Sol and Luna lies not in their pricing, but in their ability to handle agentic work at a fraction of the cost of the frontier model. OpenAI's benchmark narrative is carefully constructed, positioning Sol and Luna as close enough to the frontier to make the price difference the deciding factor.

However, this strategy may be a defensive move, as Anthropic's more capable models continue to release, leaving OpenAI to win on price rather than quality. Developers have responded to the price reduction positively, but the access limitations to Sol and the potential inefficiencies in certain workflows highlight areas of concern for the platform's usability.