🔍 Read the full analysis: Half-Price GPT‑6 Sol And Luna Models Maintain Benchmark Stability—Here’s How on ThorstenMeyerAI.com
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI launched GPT-6 Sol and Luna models at 50% lower prices, with comparable performance to previous versions. Cost reductions are achieved through improved caching and inference efficiencies, impacting AI affordability and scalability.
OpenAI has introduced GPT-6 Sol and Luna models at half the price of their GPT‑5.6 predecessors, effective immediately. The models are designed to democratize access to advanced AI by significantly reducing costs without sacrificing benchmark performance, marking a shift in how AI can be integrated into products and workflows.
Both models, launched on September 22, 2026, benefit from improvements in caching and inference technology, which allow OpenAI to lower operational costs. GPT-6 Sol is priced at $2.00 per 1 million input tokens and $10.00 per million output tokens, while GPT-6 Luna costs $0.10 and $0.50 respectively, representing a 50% reduction relative to GPT‑5.6. Despite the lower prices, independent analysis by Artificial Analysis indicates that the models maintain roughly similar benchmark scores, with Sol scoring 48 and Luna 37 on the Artificial Analysis Intelligence Index, well above median scores for their class.
Cost savings are primarily achieved through more efficient caching, with cached input reads discounted by 90%, and inference improvements that do not significantly impact model intelligence. The models also show notable reductions in hallucination rates—Sol’s hallucination rate dropped from 92% to 60%, and Luna’s from 93% to 77%—though this comes with a trade-off: Sol now attempts fewer questions, decreasing its overall answer rate from 99% to 83%. While some evaluations, particularly in knowledge work, show regressions, OpenAI emphasizes that the models’ core capabilities remain stable, making them suitable for cost-sensitive applications.
GPT‑6 Sol and Luna: half the price, about the same intelligence
OpenAI’s September 22, 2026 release doesn’t raise the ceiling. It lowers the cost of everything below it, which changes what’s worth automating.
Per 1M input / output tokens. Cached input reads keep the 90% discount.
Cost per task, halved
Measured by Artificial Analysis as the weighted cost of one Intelligence Index task, at max effort.
The effort dial moves cost more than the model choice
| Model and effort | Intelligence Index | Cost per task |
|---|---|---|
| GPT‑6 Sol (max) | 48 | $1.06 |
| GPT‑6 Sol (low) | 34 | $0.13 |
| GPT‑6 Luna (max) | 37 | $0.07 |
| GPT‑6 Luna (low) | 21 | $0.0045 |
| GPT‑6 Luna (non‑reasoning) | 18 | $0.01 |
Sol at low effort keeps about 70% of its max score for roughly an eighth of the cost, because it writes far fewer reasoning tokens. For reference, Claude Opus 5.5 leads the same index at 58.
What got better, and what got worse
Better
- Hallucination rate on AA‑Omniscience: Sol 92% → 60%, Luna 93% → 77%
- Coding Agent Index: Sol 57, up 2 points, at ~50% lower cost per task
- OpenAI reports about half as many factual mistakes for Sol as its predecessor
- Higher cache hit rates; GitHub reports over 50% fewer prompt tokens needing fresh processing
Sol gets there partly by declining more: it attempts 83% of questions vs 99%, and accuracy falls 59% → 54%.
Worse
- GDPval‑AA v2.1: Sol down ~100 Elo, Luna down ~75
- AA‑Briefcase v1.1: Luna down ~45 Elo
- Coding Agent Index: Luna 41, down 2 points
- Both models write more output tokens per task than their predecessors
Reviewers attribute the drops to weaker presentation and deliverables that omit required elements.
What to do about it
Implications for AI Deployment and Cost Efficiency
The release of GPT-6 Sol and Luna at half the previous prices significantly lowers the barrier to deploying advanced AI in a broader range of applications. For businesses and developers, this means more affordable automation of tasks such as customer support, research, and content generation, potentially accelerating AI adoption across industries. The models’ maintained benchmark performance alongside reduced costs could reshape market dynamics, encouraging more organizations to integrate AI into their workflows without prohibitive expenses.
However, the trade-offs in hallucination rates and presentation quality suggest users need to carefully evaluate whether these models meet their specific needs, especially in tasks requiring high accuracy and detailed outputs. The models’ improved efficiency in caching and inference also points to ongoing advancements in AI infrastructure, which could further reduce operational costs in the future.
Products Worth Considering
Dyson V15 Detect Cordless Vacuum
Arlo Essential Wireless Security Camera 2K (3rd Gen) – 3-Cam Kit with Yard Sign
Philips Hue Festavia 65-Foot Smart LED String Lights, 250 Color-Changing Mini LEDs
As an affiliate, we earn on qualifying purchases.
Background on Model Pricing and Performance
Prior to this release, OpenAI’s GPT‑5.6 models set the standard for high-performance AI, but their higher costs limited widespread adoption in cost-sensitive scenarios. The introduction of Astra, the top-tier model, exemplified the company’s focus on delivering the best results regardless of expense. The new GPT-6 Sol and Luna models mark a strategic shift toward making advanced AI more accessible by drastically reducing prices while maintaining comparable benchmark scores and capabilities.
Independent evaluations by Artificial Analysis, published the same day, confirm that cost per task has roughly halved, with minimal impact on overall performance metrics. The models’ ability to deliver similar or improved scores in coding and reasoning tasks, combined with lower hallucination rates, underscores the significance of these updates in the AI landscape.
Remaining Questions About Model Performance and Use Cases
While benchmark scores and hallucination rates have been evaluated, some performance regressions in knowledge work tasks suggest that model presentation quality may have declined in certain contexts. It is not yet clear how these models will perform in highly specialized or sensitive applications, or whether future updates will address these regressions. Additionally, the long-term implications of reduced answer attempts on user experience and accuracy remain to be seen.
Next Steps for Adoption and Evaluation
Organizations considering adopting GPT-6 Sol and Luna should conduct thorough testing within their specific workflows, especially for tasks requiring detailed outputs. OpenAI is expected to release further technical documentation and user guidance to help optimize model deployment. Monitoring user feedback and performance metrics will be crucial in assessing the models’ suitability for diverse applications, and future updates may refine their capabilities further.
Key Questions
How much cheaper are GPT-6 Sol and Luna compared to previous models?
GPT-6 Sol and Luna are priced at approximately 50% less than GPT‑5.6 models, with Sol at $2.00 per 1 million input tokens and $10.00 per million output tokens, and Luna at $0.10 and $0.50 respectively.
Do these models perform as well as previous versions?
Independent analysis indicates that the models maintain similar benchmark scores, with Sol scoring 48 and Luna 37 on the Artificial Analysis Intelligence Index, comparable to or above previous models. However, some knowledge-based tasks have shown regressions.
What are the main improvements in these models?
The models benefit from enhanced caching and inference techniques, which reduce operational costs. They also exhibit lower hallucination rates, making their responses more reliable, though with some trade-offs in presentation quality.
Are there any drawbacks to these models?
Yes, some evaluations have noted regressions in knowledge work and reduced presentation quality, which could affect tasks requiring detailed and well-structured outputs. Users should test thoroughly before deployment.
What should organizations do next?
Organizations should evaluate these models in their specific workflows, especially for tasks demanding high accuracy, and stay tuned for further updates and guidance from OpenAI to optimize their use.
Source: ThorstenMeyerAI.com
