OpenAI, Anthropic Both Launch Lower-Cost New Models

OpenAI released GPT-6 Sol and Luna on September 22, 2026 (local time), with API pricing cut 50% compared to GPT-5.6 promotional pricing, while Anthropic launched Claude Opus 5.5, which it says lowers operating costs by about 40% compared to Opus 5.

What did the two companies release?

• GPT-6 Sol: One tier below 'Astra,' the top model in OpenAI's GPT-6 lineup, focused on complex tasks such as coding.
• GPT-6 Luna: Designed to quickly handle high-volume tasks such as information extraction and document summarization.
• Claude Opus 5.5: The latest model in Anthropic's Claude 5.5 lineup. Anthropic said improved token efficiency lowered operating costs by about 40% compared to Opus 5.

Diane Penn, Anthropic's head of product management, research, and labs, said, "One area we keep innovating in is how to make thinking and answering more efficient," explaining that token usage can be reduced depending on how users set the thinking level.

Why has price competition intensified?

Financial News reported that, separate from calls to slow the pace of frontier model development for AI safety reasons, enterprise customers are feeling the burden of rising AI spending and demanding models that can handle the same tasks at lower cost. The successive release of relatively cheap open-weight models by Chinese firms such as Alibaba, Moonshot AI, and DeepSeek was also cited as a factor pressuring both companies to cut prices and improve token efficiency.

Implications for Korean Enterprise AX

The two figures use different baselines. OpenAI's 50% refers to API pricing versus GPT-5.6 promotional pricing, while Anthropic's roughly 40% reflects lower operating costs from improved token efficiency. When translating overseas announcements into domestic budget planning, it's safer to first distinguish whether it's a unit price cut or a reduction in tokens used for the same task.

OpenAI's split between a model for complex tasks (Sol) and one for bulk tasks (Luna) can be read as a signal to select models by task unit. If your organization is using the same model for both coding assistance and document summarization, it's worth comparing how costs change when models are assigned separately by task. Anthropic's mention of thinking-level settings follows the same logic—if you're already using the API, checking your current settings alone can be a starting point for cost assessment.

There's one thing to check this week: count the share of your organization's AI calls that consist of bulk tasks like information extraction and document summarization.

Source: "Let's Slow Down AI"... Then OpenAI and Anthropic Fire Back with Cheaper New Models