🔍 Read the full analysis: Lower Prices For GPT‑6 Sol And Luna: OpenAI Ensures Benchmark Consistency on ThorstenMeyerAI.com
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
TL;DR
OpenAI has announced a 50% price reduction for its GPT‑6 Sol and Luna models, achieved through improved caching and inference. The move aims to make advanced AI more accessible while maintaining performance benchmarks, impacting AI deployment costs significantly.
OpenAI has announced a 50% reduction in prices for its GPT‑6 Sol and Luna models, effective immediately from September 22, 2026. The new pricing aims to make advanced AI more affordable without sacrificing benchmark performance, marking a significant shift in the company’s strategy to democratize access to its models.
The new models, GPT‑6 Sol and Luna, are priced at $2.00 and $0.10 per 1 million input tokens, and $10.00 and $0.50 per 1 million output tokens, respectively. These prices are half of their GPT‑5.6 predecessors, a reduction achieved through improvements in caching and inference technology, which lower operational costs. Despite the lower prices, the models maintain comparable performance levels, with GPT‑6 Sol scoring 48 and Luna scoring 37 on the Artificial Analysis Intelligence Index, both well above their respective medians.Artificial Analysis’s evaluation, released the same day, confirms that costs per task have roughly halved, with GPT‑6 Sol’s per-task cost dropping from $1.99 to about $1.06, and Luna’s from approximately $0.17 to $0.07. These savings are due to better caching strategies, including 90% discounts on cached input reads, and more efficient inference, rather than improvements in model efficiency or capabilities. The models now support larger context windows—up to 872,000 tokens for Sol and 1 million tokens for Luna—enabling more complex tasks.
In terms of quality, GPT‑6 Sol at maximum effort has improved in hallucination reduction, with the rate dropping from 92% to 60%, and Luna from 93% to 77%. However, these models now refuse to answer a higher proportion of questions—Sol attempts 83% versus 99% for its predecessor—resulting in fewer incorrect answers but also more refusals. Some evaluations show regressions in knowledge-based tasks, with scores dropping by approximately 75–100 Elo points on certain benchmarks, attributed to changes in presentation quality and answer completeness, which may affect workflows requiring detailed outputs.
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 Cost-Effective AI Deployment
The price reductions for GPT‑6 Sol and Luna are significant because they lower the cost barrier for integrating advanced AI into products and workflows. This shift enables organizations to automate more tasks at a lower expense, expanding the scope of AI applications in customer service, research, and data analysis. The improvements in caching and inference technology demonstrate how operational efficiencies can be achieved without sacrificing model performance, potentially setting new industry standards for cost management in AI deployment.
Moreover, maintaining benchmark consistency despite lower prices reassures users about the models’ capabilities, encouraging broader adoption. The reduction in hallucinations and improved factual accuracy further enhances the models’ suitability for critical tasks, although the increased refusal rate may require adjustments in workflow design. Overall, this move could accelerate AI integration across sectors, fostering innovation and reducing operational costs.
As an affiliate, we earn on qualifying purchases.
Background on GPT‑6 Model Pricing and Performance
OpenAI’s GPT‑6 models, introduced in September 2026, marked a major step in AI development, with Astra representing the top-tier, high-performance model. Prior to the price cuts, GPT‑6 Sol and Luna were priced similarly to previous models, but with the promise of improved efficiency and larger context windows. The models’ evaluation by independent analysts, such as Artificial Analysis, showed that while costs per task decreased significantly, the models maintained or slightly improved their benchmark scores, confirming their competitive positioning.
OpenAI has emphasized that these models leverage advancements in caching and inference, which reduce operational costs. The release coincided with a broader industry trend of reducing AI costs, exemplified by competitors like Anthropic, which announced a 20% price cut for their Claude Opus 5.5. The strategic focus on cost efficiency aligns with OpenAI’s goal to democratize AI access and foster widespread adoption, especially for organizations with limited budgets.
As an affiliate, we earn on qualifying purchases.
Remaining Questions About Model Capabilities and Use Cases
It is not yet clear how these models will perform in long-term or highly specialized tasks, especially given the noted regressions in some knowledge-based evaluations. The impact of increased refusal rates on workflows requiring detailed outputs remains to be seen, and the long-term stability of cost savings through caching improvements is still under observation.
Further, the extent to which these models can replace higher-tier Astra models in complex applications without performance loss remains uncertain, as does their adaptability to different industry-specific fine-tuning needs. OpenAI has not yet disclosed detailed metrics on model robustness over extended use or in real-world deployments.
AI inference optimization hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Performance Monitoring
OpenAI is expected to continue monitoring the performance of GPT‑6 Sol and Luna in various operational contexts, gathering user feedback and refining caching strategies. The company may also release updated versions or tuning options to address the observed regressions in knowledge tasks and presentation quality.
Organizations considering integrating these models should conduct thorough testing, particularly if their workflows depend on detailed, high-quality outputs. Industry analysts anticipate that further cost reductions and performance improvements could follow as OpenAI optimizes caching and inference technologies further.
Additionally, OpenAI might expand its diagnostics and management tools, providing users with better control over effort levels and caching parameters, enabling more tailored deployment strategies.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much cheaper are GPT‑6 Sol and Luna compared to previous models?
GPT‑6 Sol’s prices are approximately 50% lower, at $2.00 per 1M input tokens and $10.00 per 1M output tokens, compared to GPT‑5.6. Luna’s prices are about 50% lower as well, at $0.10 and $0.50 respectively.
Do the new models perform as well as previous versions?
In general, the models maintain benchmark scores comparable to or slightly above previous versions, with improvements in hallucination reduction. However, some knowledge-based tasks have seen score regressions, and the models tend to refuse more questions, which may affect certain workflows.
What technological improvements enabled these price reductions?
OpenAI improved caching strategies, including 90% discounts on cached input reads, and optimized inference processes, which lower operational costs without changing the models’ core capabilities.
Will these models replace higher-tier GPT‑6 models like Astra?
While they offer strong performance at lower costs, Astra remains the top of the range for tasks requiring the best results. These models are intended to expand accessibility and cost-efficiency for broader applications.
Are there any limitations to using GPT‑6 Sol and Luna now?
Yes, some evaluations show regressions in detailed knowledge and presentation quality, and increased refusal rates may impact workflows that depend on comprehensive outputs. Users should test thoroughly before deployment.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
