OpenAI has released two new versions of its GPT-6 model, GPT-6 Sol and Luna, cutting prices by half compared to their GPT-5.6 predecessors while claiming superior performance on key benchmarks. The move comes just hours after Anthropic announced Claude Opus 5.5, marking an aggressive escalation in the AI industry's price war over who can deliver capable models most affordably.

What Are the New Pricing Tiers for GPT-6 Sol and Luna?

OpenAI announced the pricing details on its blog, emphasizing cost-efficiency gains from improvements in caching and inference optimization. GPT-6 Sol, designed for complex work including coding, now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 respectively under GPT-5.6 promotional pricing. GPT-6 Luna, the smaller model intended for high-volume tasks like document summarization and short-answer questions, dropped to $0.10 and $0.50 per million tokens, from $0.20 and $1.20.

OpenAI also increased its default cache hit rates, reducing cached input-token reads by 90 percent. This technical improvement allows developers to set explicit breakpoints in prompts and adjust reasoning effort without losing cached context, further reducing costs per task.

How Do Sol and Luna Compare to Anthropic's New Models?

The timing of these releases reveals the intensity of competition in the AI market. Anthropic launched Claude Opus 5.5 approximately 90 minutes before OpenAI's announcement, claiming 40 percent cost reductions compared to Opus 5 while maintaining performance close to Claude Fable 5.1. Both companies are pursuing the same strategy: taking existing capabilities and delivering them at lower prices rather than introducing frontier models with new abilities.

OpenAI's benchmarks directly target Anthropic's offerings. On the AutomationBench test of business workflows, GPT-6 Sol at maximum effort scored 33.2 percent at $0.27 per task, compared to what OpenAI claims is Claude Opus 5 at 26.9 percent and 11.1 times the cost. On the Agents' Last Exam evaluation, Sol achieved 56.4 percent, reportedly beating Opus 5's best result at 60 percent lower cost per task. On the DeepSWE software engineering benchmark, Sol scored 68.8 percent against Claude Fable 5's 69.9 percent, but at roughly 80 percent less cost per task.

OpenAI acknowledged caveats in its comparisons, noting that competitor scores came from published reports rather than in-house testing, and that some Fable 5.1 results were substituted with Fable 5 data where unavailable.

Why Is This Price War Happening Now?

The pressure driving these price cuts comes from multiple directions. Chinese open-weight models from companies like Alibaba, DeepSeek, and Moonshot now perform much of this work for free, forcing both OpenAI and Anthropic to compete on cost. Startups have increasingly migrated to cheaper open-source alternatives as their AI bills have mounted. Both labs have also publicly called for the industry to slow frontier model development, and these releases represent their first major announcements since making those calls.

"They are also half the price per token, and even less per task," Sam Altman wrote on X about the new models.

Sam Altman, CEO at OpenAI

Ara Kharazian, lead economist at the spend management company Ramp, told Fortune that the two companies are engaged in a price war that is driving down the price of AI itself, along with their ability to profit from it. The pressure is being fought on two fronts: through cheaper models and through outright cuts on expensive ones.

How to Evaluate Which Model Fits Your Use Case

• Complex Tasks: GPT-6 Sol is built for sophisticated work including coding, software engineering, and multi-step reasoning, making it suitable for developers and technical teams who need strong performance without paying for the most expensive tier.
• High-Volume Operations: GPT-6 Luna is designed for repetitive, goal-oriented tasks such as summarizing documents, answering frequently asked questions, and processing large batches of similar requests at minimal cost.
• Cost Optimization: Both models are now available in ChatGPT Work and Codex across Plus, Pro, Business, Enterprise, and Edu accounts, with Luna also available to Free and Go users on the desktop app, allowing organizations to choose based on their budget and performance requirements.

The rollout is gradual to maintain service stability. In the API, the models are called gpt-6-sol and gpt-6-luna. Neither is yet available in Chat, though OpenAI said it would expand availability through the day.

What Do These Models Claim About Reliability and Safety?

OpenAI claims improved factual reliability in both models. Using an internal test based on de-identified ChatGPT conversations where users flagged errors from earlier models, OpenAI says Sol makes about half as many mistakes as GPT-5.6 Sol. Luna at higher effort levels matches GPT-5.6 Sol's accuracy at roughly one-hundredth of the cost. The company notes these conversations were specifically chosen for being error-prone and do not represent typical usage patterns.

On alignment and safety, both models score better than their GPT-5.6 counterparts across OpenAI's internal tests, including a lower rate of misleading claims about their own coding capabilities. Full results are available in OpenAI's system card documentation.

The broader context matters here. OpenAI recently announced it would allow outside groups to run technical safety evaluations during model training rather than only before release, a shift toward more transparent development practices.

What Does This Mean for the AI Industry?

These releases underscore a fundamental shift in AI competition. The race is no longer primarily about who builds the most capable frontier model, but rather who can deliver sufficient capability at the lowest cost. This dynamic mirrors earlier technology cycles where commoditization follows rapid innovation. For users and organizations, it means AI tools are becoming more accessible and affordable, though it also signals that profit margins in AI services may continue to compress as companies compete on price rather than exclusive capabilities.