OpenAI Halves GPT‑6 Sol And Luna Prices And Keeps Benchmark Scores Stable

🔍 Read the full analysis: OpenAI Halves GPT‑6 Sol And Luna Prices And Keeps Benchmark Scores Stable on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get the latest gadgets delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OpenAI has announced a significant reduction in the prices of its GPT-6 Sol and GPT-6 Luna models, slashing costs by 50% while maintaining their benchmark scores. The move, revealed on September 22, 2026, aims to broaden access to advanced AI capabilities by making mid-tier models more affordable for developers and enterprises. Despite the lower prices, the models’ performance metrics, including benchmark scores and inference speeds, remain stable, thanks to improvements in caching and inference technology.

On September 22, 2026, OpenAI introduced GPT-6 Sol and GPT-6 Luna as part of its Astra family, with prices cut by half compared to their GPT‑5.6 predecessors. The pricing now stands at $2.00 per 1 million tokens for input and $10.00 for output on Sol, and $0.10 and $0.50 respectively for Luna, representing a 50% reduction. These changes were enabled by advancements in caching and inference, which allow the models to operate at lower costs while passing savings to customers.

Independent analysis from Artificial Analysis confirmed that while the costs per task have halved—$1.06 for Sol and $0.07 for Luna at maximum effort—the benchmark scores remained roughly stable. Sol scored 48 on the Artificial Analysis Intelligence Index, well above the median of 25, with Luna scoring 37 against a median of 12. These scores indicate that the models’ core capabilities are preserved despite the price cuts.

OpenAI reports that the models’ hallucination rates have improved significantly, with Sol’s hallucination rate dropping from 92% to 60%, and Luna’s from 93% to 77%. This reduction is partly achieved by the models declining to answer more often, which lowers errors but also reduces their willingness to respond, a trade that may impact certain workflows. The models also show some regression in knowledge tasks, with decreased performance on specific benchmarks, attributed to changes in presentation quality and output length.

Impact on AI Accessibility and Cost Efficiency

The price reductions for GPT-6 Sol and Luna could make advanced AI more accessible to a wider range of users, including startups and smaller enterprises that previously found the costs prohibitive. Lower costs for mid-tier models expand the scope of automation and AI integration in business workflows, potentially accelerating AI adoption across industries. Maintaining benchmark scores while reducing prices demonstrates that OpenAI is prioritizing cost efficiency without sacrificing core model capabilities, which could influence competitive dynamics in AI markets.

Furthermore, the improvements in hallucination rates and caching efficiency suggest that the models are becoming more reliable and cost-effective for tasks involving customer support, research, and data extraction. However, the noted regressions in some knowledge benchmarks highlight that not all performance aspects are improved, and users may need to test models for specific use cases.

Background on OpenAI’s Model Pricing and Performance

OpenAI’s recent model releases have consistently aimed to balance performance and cost. The GPT‑6 Astra family, introduced earlier in September 2026, marked a step toward making high-performance models more affordable. The pricing for previous models, such as GPT‑5.6, was significantly higher, limiting accessibility for some users. The new models, Sol and Luna, are positioned as more cost-effective options within the same family, leveraging technological improvements in caching, inference, and prompt management to reduce operational costs.

Independent evaluations, such as those from Artificial Analysis, have historically tracked the performance of OpenAI’s models across various benchmarks, including intelligence, coding, and knowledge tasks. These assessments provide a baseline for understanding how recent price cuts impact model quality and utility. Prior to this release, OpenAI had emphasized that improvements in caching and inference would enable lower costs without degrading performance, a claim now supported by independent data.

This development follows a broader industry trend of reducing AI model costs to democratize access and foster innovation, with competitors like Anthropic also cutting prices on similar models. The balance between cost and performance remains a key concern for users deploying AI at scale.

Unresolved Aspects of Model Performance and Use

It is not yet clear how the reductions in hallucination rates and the stability in benchmark scores will translate to real-world applications across diverse industries. The noted regressions in knowledge benchmarks suggest potential trade-offs that may impact specific workflows, especially those requiring detailed and comprehensive outputs. Additionally, the long-term effects of increased refusal rates on user experience and task completion are still to be observed.

Further, the extent to which these models will continue to improve in other performance areas remains uncertain, as OpenAI has not disclosed detailed updates on training data or tuning strategies that might influence future capabilities.

Next Steps for Adoption and Evaluation

OpenAI is expected to release more detailed technical documentation and user guidelines to help organizations evaluate the models for their specific use cases. Industry analysts and early adopters will likely conduct further testing to confirm the models’ performance in real-world scenarios, especially in customer-facing and knowledge-intensive tasks.

Monitoring how these models perform over time, especially regarding hallucination rates, knowledge accuracy, and output quality, will be critical. OpenAI may also introduce further updates or new models aimed at addressing current regressions or expanding capabilities.

Overall, the focus will be on balancing cost efficiency with reliability, with the potential for these models to reshape AI deployment strategies in the near future.

Key Questions

How much cheaper are GPT‑6 Sol and Luna models now?

GPT‑6 Sol is now priced at $2.00 per 1 million tokens for input and $10.00 for output, roughly half the previous GPT‑5.6 prices. Luna costs $0.10 and $0.50 respectively, representing approximately a 50% reduction.

Do the models perform as well as before?

Benchmark scores indicate that core capabilities remain stable, with Sol scoring 48 and Luna 37 on the Artificial Analysis Intelligence Index, both above median levels. However, some knowledge benchmarks showed regressions, and hallucination rates have improved but still vary.

What technological improvements enabled the price cuts?

OpenAI improved caching and inference processes, allowing the models to operate at lower costs while maintaining performance. These include higher cache read discounts and better prompt reuse strategies.

Are there any drawbacks to the new models?

Yes, some regressions in knowledge benchmarks and increased refusal rates may impact workflows requiring detailed or comprehensive outputs. Users should test models for their specific needs before full deployment.

What could be the next developments from OpenAI?

OpenAI is likely to release further updates, improve model performance, and expand capabilities while continuing to focus on reducing costs and increasing reliability for diverse applications.

Source: ThorstenMeyerAI.com