OpenAI has released two new AI models, GPT-6 Soul and GPT-6 Luna, that dramatically reduce costs while improving accuracy compared to previous versions. The launch marks OpenAI's latest move to make advanced AI more affordable and reliable for everyday professional work, from coding to document summarization. Soul and Luna arrive just 90 minutes after Anthropic announced its own efficiency-focused Opus 5.5 model, signaling an intensifying race between the two AI giants to deliver smarter, cheaper systems.

What Are GPT-6 Soul and Luna Designed to Do?

The two models serve different purposes within OpenAI's expanding GPT-6 family. Soul is built for complex, reasoning-heavy tasks like coding and advanced professional work that requires deep analysis. Luna, by contrast, targets high-volume office tasks with straightforward goals, such as summarizing documents, extracting key information, and answering routine questions. Both models use training methods similar to GPT-6 Astra, OpenAI's flagship model released earlier in September, but are optimized for speed and cost-efficiency rather than maximum capability.

How Much Cheaper Are These Models Than Previous Versions?

The pricing improvements are substantial. GPT-6 Soul costs $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna costs just $0.10 per million input tokens and $0.50 per million output tokens. This represents roughly a 50% reduction in API costs compared to GPT-5.6 versions of both models. OpenAI attributed the cost savings to improvements in caching and inference technologies, which allow the models to process information more efficiently without sacrificing quality.

To put this in practical terms, Luna's pricing means organizations can run high-volume automation tasks at a fraction of previous costs. According to OpenAI's internal testing, GPT-6 Luna outperformed its GPT-5.6 predecessor by 5.4% on automation-task performance while reducing the cost per task by 58%.

Where Do These Models Rank in Accuracy and Error Rates?

Accuracy improvements are equally impressive. In internal evaluations based on real-world user conversations where errors had been reported, GPT-6 Soul made approximately half as many errors as its predecessor. OpenAI stated that Soul now offers reliability close to Astra, the company's most powerful model, but at significantly lower cost. This is a critical advantage for organizations that need both accuracy and affordability.

Luna also showed measurable gains in factual accuracy and coding reliability. The company presented results from multiple performance tests showing both new models outperforming previous versions across the evaluations OpenAI conducted. These improvements matter because they reduce the need for human review and correction, saving time and money for users.

How to Access GPT-6 Soul and Luna

• ChatGPT Work and Codex: Available to users on Plus, Pro, Enterprise, Business, and Edu subscription plans through ChatGPT's dedicated work interface and coding tools.
• OpenAI API: Developers can integrate Soul and Luna directly into applications and services via OpenAI's application programming interface for custom implementations.
• Desktop App: Free and Go plan users can access Luna through ChatGPT's desktop application, though Soul remains limited to paid tiers.

At the time of announcement, the models were not yet available in ChatGPT's main chat interface, but OpenAI said it would gradually roll them out through the app and website throughout the day.

How Does This Compare to Anthropic's Latest Model?

The timing of these launches reveals the intensity of competition between OpenAI and Anthropic. Anthropic announced its Opus 5.5 model approximately 90 minutes before OpenAI unveiled Soul and Luna. OpenAI claims its new models outperform Anthropic's advanced models, including Fable and Opus, on several tasks based on performance tests published by the company. However, both companies are pursuing similar strategies: delivering more efficient, accurate models at lower costs to capture market share in the rapidly growing AI application space.

This competitive pressure is beneficial for users and developers, as it drives both companies to improve performance while reducing prices. The race to deliver cheaper, more accurate models reflects a broader industry trend toward making AI tools accessible to smaller organizations and individual developers who previously could not afford enterprise-grade systems.

The launch of GPT-6 Soul and Luna comes only a few months after OpenAI released GPT-5.6 versions of the same models, demonstrating the company's rapid iteration cycle. In a relatively short period, OpenAI has significantly reduced error rates and operating costs, suggesting that the pace of AI improvement shows no signs of slowing.