Anthropic and OpenAI on September 22 unveiled their new AI models; models that provide some capabilities of these companies' flagship models to users at lower cost. This simultaneous release comes as debate over the pace of AI development and the risks posed by advanced models has once again drawn attention.
Anthropic introduced Claude Opus 5.5 that day; a model which, according to the company, approaches the performance level of Claude Fable 5.1, the company's more advanced model, on most tasks and costs about 40 percent less to run than Opus 5. Anthropic set the price of Opus 5.5 at $4 per one million input tokens and $20 per one million output tokens.
While reducing cost, this model is designed for complex tasks including programming, professional tasks, and multi-step work. Anthropic says one early tester was able to migrate a project of about 680,000 lines of code in less than a day using Opus 5.5; work that, the company says, could have taken an engineering team weeks.
On the other side, OpenAI also released GPT-6 Sol and GPT-6 Luna. These two models are built on advances made in GPT-6 Astra, but are intended to offer better suitability for everyday and high-volume use, with different cost and speed profiles. According to OpenAI, Sol is designed to balance capability and cost, while Luna is a cheaper option for high-volume, cost-sensitive tasks.
The official price of GPT-6 Sol is $2 per one million input tokens and $10 per one million output tokens, while GPT-6 Luna costs $0.10 and $0.50 respectively. OpenAI said these models were developed using more efficient caching and execution methods and that the processing cost reductions have been passed on to users.
One notable point in the release of Opus 5.5 is Anthropic's focus on safety. The company said the new model was tested before release by external evaluators, including Frontier Design and METR, and carries the same protective measures developed for Anthropic's more powerful models. These measures are particularly important in areas such as cybersecurity and biology.
Anthropic also said that Opus 5.5 achieved the best performance to date in the company's internal comprehensive behavioral and alignment evaluation. However, safety evaluations remain an important part of the development process for new models, because the behavior of powerful models in real-world conditions is not always fully predictable from controlled tests.
These developments occur as both companies face increasing pressure to improve the economic efficiency of their models. In addition to direct competition among American companies, low-cost and open-source models developed in China have added further pressure to the market. Therefore, reducing the cost of running models can be as important as increasing their raw capabilities for attracting users and enterprise customers.
The release of Opus 5.5 also comes at a time when Anthropic is preparing to enter the public capital markets. published reports have cited the company's initial public offering as one of the major upcoming events in the AI industry; meanwhile, OpenAI is also exploring funding paths and its corporate structure.
Finally, the simultaneous release of these models shows that the next-generation AI competition is moving toward building models that are not only powerful but also faster, cheaper, and usable at large scale. For companies spending billions to train and run AI models, increased efficiency and lower cost per request can play a decisive role in expanding commercial adoption of this technology.