VentureBeat reports that OpenAI is expanding its range of AI models with the introduction of new models GPT-6 Sol and GPT-6 Luna. These mid- and low-end models are designed to perform common tasks that require less brainpower, and cost half or less than their predecessors when accessed through the OpenAI API. Both models outperform their GPT-5.6 predecessors, although they remain less efficient and powerful than the flagship GPT-6 Astra model launched earlier this month. The GPT-6 Luna model costs only $0.10/$0.50 for 1 million tokens, and the GPT-6 Sol costs $2/$10 for 1 million tokens. OpenAI says improvements in processing and caching have lowered prices while increasing capabilities. An OpenAI spokesperson confirmed to VentureBeat that the prices listed for GPT-6 Sol and Luna are permanent promotions and not temporary promotions. Compared to current prices for GPT-5.6, the price reduction for GPT-6 Sol is 50% in both directions, and for Luna - 50% in entry and 58.3% in exit. GPT-6 Sol is focused on complex tasks such as feature development, code review and data analysis, while Luna is designed for you
large-scale work, such as summarizing and extracting information. Astra remains for the most complex and multifaceted projects. Prices for the GPT-6 Sol are comparable to those of the Claude Sonnet 5, making it competitive in this niche. At the same time, Luna occupies its own price segment, offering significantly lower token rates than other models such as Sonnet 5 and Gemini 3.8 Flash. OpenAI also emphasizes that the cost per successful completion of a task becomes the main criterion for evaluating models. For example, on AutomationBench 1.0.6 GPT-6 Sol achieved a 33.2% success rate at a cost of $0.27 per task, which is significantly lower than its competitors. The GPT-6 Sol and Luna models will be available through the OpenAI API, as well as ChatGPT Work and Codex for Plus, Pro, Business and Enterprise customers. Moreover, OpenAI argues that the continued decline in prices for these models strengthens the ability to choose between different classes of work in a portfolio of AI models. The material was prepared by a neural network based on the original publication. AI may make mistakes in details (numbers, names, dates, quotes) - for important information, check with the original source. Anthropic presented Claude Opus 5.
5, beating Fable 5.1 in key indicators Qualcomm introduced new processors for smartphones with an emphasis on AI Kyutai released a Voice of Reason model for verbally solving mathematical problems. Nokia opens access to AnyJev, a library for decision-making without training. OpenAI introduced GPT-6 Sol and Luna with new API pricing, reducing the cost by 50%.
