Claude Opus 5.5 and GPT-6 Sol and Luna released
A look at the new Anthropic and OpenAI models: lower usage costs, changes to writing style and availability in apps.
Updated:
On September 22, Anthropic introduced Claude Opus 5.5 and OpenAI released GPT-6 Sol and Luna. Both companies report improvements in work tasks and lower usage costs compared with their previous models.[1][2]
Claude Opus 5.5
Source: Anthropic.
According to Anthropic’s measurements, compared with Opus 5, Opus 5.5 costs around 40 percent less on typical workloads and generates text more than 30 percent faster. The saving refers to using the model to complete tasks, not a 40 percent reduction in the price of a Claude subscription.[1]
Anthropic has also changed the writing style. It says Opus 5.5 puts important information first, uses less jargon and follows style instructions more closely.[1]
The company also highlights improvements in research, financial analysis and business materials. In one internal test, Opus 5.5 analysed a fictional merger, built a financial model and prepared a presentation faster and at lower cost than its predecessor.[1] The result applies to this test; savings on other tasks may differ.
GPT-6 Sol and Luna
Source: OpenAI.
OpenAI is expanding the GPT-6 family with Sol and Luna. Astra remains its most capable model, while the new variants are designed for faster, more affordable use.[2]
Sol is intended for more demanding everyday work, while Luna is the cheaper variant. OpenAI reports improvements in both models across work with applications, coding and computer use.[2]
According to OpenAI, Sol makes roughly half as many mistakes as its predecessor in its internal factuality evaluation. However, that evaluation uses deliberately selected conversations in which users had already flagged errors, rather than measuring all ordinary responses.[2]
OpenAI also emphasises clearer language, fewer unnecessary details and slightly shorter answers.[2]
Pricing and availability
The biggest price reductions in these announcements concern programmatic access to models, or the API, through which companies integrate them into applications and automations. For example, OpenAI has halved Sol’s input and output text prices compared with GPT-5.6 promotional pricing.[2]
At launch, Anthropic is raising five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans.[1] OpenAI is introducing Sol and Luna in ChatGPT Work and Codex, as well as through the API. According to the launch announcement, they are not yet available in the regular Chat mode.[2]
Limits of the comparison
The announcements report results from different tests, so they do not provide a reliable comparison of all three new models on the same tasks.[1][2]
Before switching models, it is worth repeating a few tasks you already perform with your current one. When comparing costs, include the time you spend checking facts and correcting responses.