GPT-6 Sol and Luna Launch: OpenAI Boosts AI Accessibility with Cost-Effective Models

OpenAI has significantly expanded its advanced AI offerings by introducing two new models, GPT-6 Sol and GPT-6 Luna, designed to democratize access to frontier intelligence. While the flagship GPT-6 Astra model remains the pinnacle of performance, reserved for the most demanding and critical projects, the new Sol and Luna models are engineered to distribute the benefits of this state-of-the-art intelligence across a wider spectrum of use cases, budgets, and operational scales. This strategic expansion directly addresses the need for powerful AI capabilities that are also highly cost-efficient and practical for everyday, large-scale applications.

The core innovation behind GPT-6 Sol and GPT-6 Luna lies in their ability to maintain the advanced performance characteristics pioneered by GPT-6 Astra—including superior professional work handling, high factuality, advanced coding proficiency, seamless computer interaction, and robust alignment—while achieving a substantial leap in affordability and speed. OpenAI notes that these models lead across the 'cost–intelligence curve,' meaning they provide exceptional, high-tier capabilities at every operational level. This efficiency is achieved through significant infrastructure improvements, specifically advancements in caching and inference techniques, which allow the models to be served at a much lower operational cost. Crucially, OpenAI is passing these savings directly to its user base by slashing the API prices for both Sol and Luna by 50% compared to the previous promotional pricing of GPT-5.6. This massive reduction in cost makes advanced AI functionality viable for a much broader range of routine tasks and enterprise-level deployments.

GPT-6 Astra is positioned as the ultimate choice, maintaining its status as the best model across all metrics for users who require uncompromising, best-in-class results. Conversely, GPT-6 Sol and GPT-6 Luna are designed to bring intelligence upgrades and cost efficiency to the models that developers and businesses already rely on for complex, day-to-day operational tasks. GPT-6 Sol, for instance, is highlighted as a powerful tool for tackling difficult work assignments while simultaneously offering users greater freedom to iterate due to higher usage limits and significantly lower costs, all while delivering superior intelligence and results compared to similarly priced competing models.

To substantiate these claims, OpenAI provided rigorous benchmark comparisons. In the AutomationBench, which tests AI agents on complex, end-to-end business workflows utilizing 47 different tools spanning critical departments like sales, marketing, operations, finance, and HR, GPT-6 Sol demonstrated remarkable efficiency. Specifically, at an 'xhigh effort' setting, GPT-6 Sol outperformed Claude Opus 5 at its 'max effort' setting, achieving this performance at a mere 9% of the cost per task. Furthermore, GPT-6 Luna showed substantial improvement over its predecessor, boosting its performance by 5.4 percentage points at a cost that was 58% lower per task. These benchmarks underscore the model's ability to deliver enterprise-grade performance without the prohibitive cost associated with top-tier competitors.

The evaluation of AI agents on complex professional workflows is further demonstrated by the 'Agents’ Last Exam.' Here, GPT-6 Sol, operating at maximum effort, scored 56.4%, significantly surpassing Claude Opus 5's highest recorded score of 60% lower cost per task. These tests are crucial because they evaluate 'long-horizon, economically valuable tasks' across 55 sub-industries, simulating the vast majority of professional work performed on a computer. The ability of Sol and Luna to maintain high scores while drastically reducing operational expenditure is a major market differentiator.

Beyond general workflow automation, the models show marked improvements in factual reliability. OpenAI emphasizes that the usefulness of any AI answer hinges on its factual accuracy. In internal factuality evaluations, which analyze real-world conversations where users have flagged factual errors, GPT-6 Sol was found to make approximately half the number of mistakes compared to its predecessor. This places its reliability close to the level of GPT-6 Astra, but at a fraction of the cost. GPT-6 Luna also shows substantial gains here, matching the performance of GPT-5.6 Sol at roughly one-hundredth of the cost at higher effort levels.

Another critical area of focus is coding. As coding agents take on tasks of increasing complexity, scope, and duration—a trend reflected by OpenAI's own internal usage, where median daily token usage for researchers exceeded $600 and the 90th percentile reached $7,000—the cost of sustained, high-quality use becomes paramount. GPT-6 Sol and Luna combine strong coding performance with dramatically lower API prices. On the FrontierCode benchmark, which grades code not only on correctness but also on 'mergeability' (including test quality, scope discipline, and adherence to codebase standards), GPT-6 Sol significantly improves upon GPT-5.6 Sol and is capable of matching Claude Fable 5.1 at the 'xhigh' setting, all while maintaining a much lower cost. The combination of high-level, professional performance with unprecedented cost efficiency makes GPT-6 Sol and GPT-6 Luna revolutionary tools for scaling AI adoption across diverse industries. This tiered approach allows businesses to select the perfect balance between performance and budget, making advanced AI accessible to a much wider user base. The measurable performance gains across multiple industry-specific benchmarks, especially when compared to high-cost competitors, solidify the models' position as industry leaders in value.