Like Anthropic, OpenAI is releasing more cost-competitive models while calling for a slowdown in frontier AI development.Author: Li Dan
Source: Wall Street Journal
The GPT-6 model family is accelerating its expansion.
On Tuesday, September 22, Eastern Time, OpenAI announced the launch of two new models: GPT-6 Sol and GPT-6 Luna. Less than three weeks after releasing its flagship model GPT-6 Astra, touted as the "smartest in the world," OpenAI has further expanded its GPT-6 product line, bringing some of Astra’s capabilities into faster and more cost-effective models.
Another key highlight of these two models is pricing: OpenAI announced that the API prices for GPT-6 Sol and GPT-6 Luna are 50% lower than the promotional pricing of GPT-5.6. OpenAI attributes this to further improvements in caching and inference efficiency, passing the cost savings directly on to users.
According to OpenAI's announcement, Sol and Luna are now available for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, with API access also enabled; free and Go users can experience GPT-6 Luna in the desktop app. However, neither model is currently available on the standard ChatGPT chat interface.
Astra was released less than three weeks ago, and OpenAI has now added two more GPT-6 models.
Looking at the release schedule of GPT-6, OpenAI has been remarkably active this time.
On September 3, OpenAI officially launched GPT-6 Astra, claiming it reaches new frontiers in capabilities across computer and browser operations, software engineering, cybersecurity, science, and professional work. At the time, OpenAI positioned Astra as the flagship model of the GPT-6 family.
Less than three weeks ago, OpenAI launched Sol and Luna again.
OpenAI's official account on social media stated that Sol and Luna are "built upon the technological advancements behind GPT-6 Astra," bringing many of Astra's advantages into "faster, more affordable models" to support scaled workloads.
This product rollout shows that OpenAI is not positioning GPT-6 as a single-point upgrade limited to a flagship model, but rather rapidly expanding its product line across different performance levels, costs, and use cases following the release of Astra.
This also aligns with the increasingly clear product strategy of AI model providers: flagship models push the boundaries of capability, while faster, more affordable versions target higher-frequency workflows.
From Astra to Sol and Luna, capabilities are being extended to larger-scale scenarios.
Unlike Astra, which emphasizes cutting-edge capabilities, Sol and Luna are positioned to highlight speed, cost-efficiency, and scalable adoption.
OpenAI stated that Sol and Luna not only build on Astra’s technological advancements but also feature optimizations for caching and inference efficiency. The company said these improvements reduce model operating costs, ultimately passing on lower API prices to users.
OpenAI also stated that both models have incorporated advancements from Astra in alignment capabilities, improving upon their respective GPT-5.6 versions.
In other words, OpenAI is not merely launching two “lower-cost” models, but rather attempting to extend some of Astra’s advancements in capability and alignment to higher-frequency, larger-scale model usage scenarios.
For developers and enterprises, this means that advanced model capabilities, previously only worthwhile to invoke for high-value tasks, may soon become available at lower unit costs for a wide range of repetitive tasks such as coding, agents, and enterprise automation.
The API price has been halved compared to the GPT-5.6 promotional rate.
Price is the clearest business signal of this release.
OpenAI announced that the API prices for GPT-6 Sol and GPT-6 Luna are reduced by 50% compared to the promotional price of GPT-5.6.
Currently, OpenAI's announced promotional price for GPT-5.6 Sol is $4 per 1 million tokens for input and $20 for output; thus, this price reduction for Sol and Luna signifies further downward pressure on their API costs. OpenAI’s enterprise pricing table also indicates that GPT-5.6 Sol is still in its promotional pricing phase.
OpenAI explained that the price reduction is primarily due to more efficient caching and inference, rather than mere business concessions.
The official OpenAI account stated that “higher usage limits and lower costs” will give users more flexibility and room for iteration.
For developers and enterprises requiring frequent model calls, a reduction in cost per token means more tasks can be run within the same budget, and agent workflows previously limited by cost may now have greater operational capacity.
This has further extended the competition among AI models from “which model is more capable” to who can provide sufficiently capable models at a lower cost.
First, enter Work and Codex; free users can experience Luna.
In terms of product coverage, OpenAI did not immediately integrate Sol and Luna fully into the general ChatGPT chat interface.
According to OpenAI’s arrangements, Plus, Pro, Business, Enterprise, and Edu users can access both models in ChatGPT Work and Codex, as well as via API calls.
Free and Go plan users can experience GPT-6 Luna on the desktop app.
This arrangement clearly positions the launch scenarios for both models toward work and development, rather than purely consumer-oriented chatting.
Among these, Sol and Luna's low cost and high efficiency make them better suited for coding tasks, agent workflows, and enterprise internal automation that require frequent model invocation.
In comparison, GPT-6 Astra still serves as the flagship model. Following Astra's release, OpenAI temporarily paused new Pro subscription sign-ups due to surging demand, describing Astra's demand as "unprecedented."
Therefore, the introduction of Sol and Luna may also help OpenAI direct some of its high-frequency demand toward lower-cost models, reducing infrastructure pressure while scaling usage.
AI model competition has shifted from competing on flagship models to competing on scale.
The release of Sol and Luna occurred against the backdrop of AI model providers aggressively launching more affordable versions.
Also on the same Tuesday, Anthropic, OpenAI’s rival, launched its new model, Claude Opus 5.5, emphasizing lower operational costs. The two AI leaders have nearly simultaneously rolled out new models with improved cost competitiveness. The focus of market competition is now shifting further toward balancing performance, speed, and price, beyond merely pursuing the upper limits of model capability.
For OpenAI, GPT-6 Astra sets the upper limit of capabilities, while Sol and Luna are tasked with extending these technological advancements to broader use cases.
From the release of Astra on September 3 to the launch of Sol and Luna less than three weeks later, the GPT-6 family is rapidly shifting from “flagship model releases” to “product line expansion.”
Meanwhile, the API price, further reduced by 50% from the GPT-5.6 promotional rate, sends a clear signal: as model capabilities gradually become homogeneous, OpenAI must not only compete for the title of the strongest model, but also for the scalable market of how many times developers and enterprises are truly willing to call the model each day.