OpenAI launches GPT-5.6: the Sol, Terra and Luna family promises more intelligence per token and prices from one dollar
OpenAI has announced the general availability of GPT-5.6, a new family of models that arrives with a clear promise: more intelligence for every token processed and better performance per dollar invested. The company led by Sam Altman structures this generation into three tiers —Sol, Terra and Luna— that coexist as permanent models and evolve independently, a change of strategy compared to the classic succession of numbered versions. The launch, already available in ChatGPT, Codex and the API, starts with prices of 1 dollar per million input tokens, placing the new generation in a particularly aggressive cost range.
Sol, Terra and Luna: a hierarchy designed for each use
GPT-5.6 Sol is the flagship model of the family and the one that concentrates the greatest improvements in programming, knowledge work, cybersecurity and science. According to OpenAI, it sets a new efficiency standard compared to previous generations and to the main competitors: it achieves comparable or superior results while consuming fewer tokens, which translates into a lower estimated total cost for companies operating at scale.
Below Sol sit Terra and Luna. Terra is presented as the balanced option for daily work, with a balance between capacity and cost designed to be integrated into production workflows. Luna, for its part, aims directly at speed and reduced price, becoming the most accessible alternative for developers and applications with a high volume of requests. The nomenclature itself, inspired by celestial bodies, reflects the new product strategy: durable tiers that can be updated over time without needing to rename the whole family.
Ultra mode: parallel agents
One of the most striking announcements is the ultra configuration, designed to coordinate multiple agents in parallel. This mode expands capacity on demand for complex tasks and allows several agents to work simultaneously on the same problem, speeding up processes that previously had to be executed sequentially. It is a direct commitment to the paradigm of autonomous agents, one of the fronts on which OpenAI wants to stand out against rivals such as Anthropic or Google DeepMind.
The company has also confirmed that GPT-5.6 Sol will arrive on Cerebras at the end of July, with a speed of up to 750 tokens per second. That deployment, however, will begin limited to selected customers while capacity is expanded, a rollout strategy that has become common in the sector.
Prices and availability
The price range of the GPT-5.6 family runs from 1 dollar per million input tokens up to 30 dollars, depending on the model and the type of use. In ChatGPT, users of the Plus, Pro, Business and Enterprise plans can access GPT-5.6 Sol through medium and high effort configurations, while Pro and Enterprise customers also have GPT-5.6 Sol Pro for especially demanding tasks.
For developers, availability in the API and in Codex opens the door to integrating the new generation into assisted programming tools, data analysis and process automation. OpenAI maintains that the improvements in programming and cybersecurity are especially relevant, and the first published benchmarks point to a notable leap in exploitation and vulnerability analysis tasks, a field in which the company has invested decisively over the last few months.
A launch conditioned by politics
The arrival of GPT-5.6 has not been without controversy. The U.S. administration pressured OpenAI to stagger the launch, and initial access was limited to a small group of government-approved partners. Sam Altman described that request as «bad news», although he acknowledged that the limited preview is «quite reasonable». The White House, meanwhile, is working with the big AI companies on a voluntary framework that would give the Government early access to cutting-edge models up to 30 days before their launch.
What it means for the sector
OpenAI’s move reinforces a trend that dominates the industry in 2026: the battle is no longer only about the raw capacity of the models, but about economic efficiency. With entry prices of one dollar per million tokens, the company pushes down the real cost of automating specialized work and opens the door for SMEs to incorporate advanced AI into their processes without needing large budgets.
It remains to be seen how competitors respond, but one thing seems clear: frontier artificial intelligence is no longer measured only by its results in benchmarks, but by how much it costs to obtain them. And in that equation, GPT-5.6 has set the bar very high.
