🔍 Read the full analysis: OpenAI Cuts Costs For GPT‑6 Sol And Luna While Benchmark Results Stay Consistent 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 announced a 50% price reduction for GPT‑6 Sol and Luna models while benchmark results remain consistent. The change aims to improve AI affordability without sacrificing performance, impacting AI deployment strategies.
OpenAI has announced a significant reduction in the costs of its GPT‑6 Sol and Luna models, slashing prices by approximately 50% while maintaining comparable benchmark performance. The move aims to expand AI deployment by making these models more affordable for a broader range of users and applications, without sacrificing the quality of results.
On September 22, 2026, OpenAI released GPT‑6 Sol and GPT‑6 Luna, two new models positioned as more cost-efficient alternatives within the GPT‑6 family. Both models are now priced at half the previous GPT‑5.6 models, with Sol at $2.00 per 1 million input tokens and $10 per 1 million output tokens, and Luna at $0.10 and $0.50 respectively. The company attributes the cost reductions to improvements in caching and inference technologies, which enable lower operational expenses while passing savings to customers.
Independent analysis from Artificial Analysis confirms that, despite the lower prices, benchmark scores for these models remain steady. Sol scores 48 on the Artificial Analysis Intelligence Index, well above the median in its class, with Luna scoring 37, also above average. Costs per task have been cut roughly in half compared to previous models, with Sol at about $1.06 per task and Luna at approximately $0.07, even as both models handle more output tokens per task. The models also demonstrate improvements in hallucination reduction, with Sol cutting hallucination rates from 92% to 60%, and Luna from 93% to 77%, primarily through increased refusal to answer uncertain questions.
However, some regressions were noted in knowledge-based evaluations, where both models scored lower in specific economic and knowledge tasks, attributed to changes in output presentation quality. OpenAI’s internal notes suggest these tuning adjustments favor more concise, less detailed responses, which may impact tasks requiring comprehensive deliverables. Effort levels and caching improvements are also highlighted as key factors in operational cost savings and model performance tuning.
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
Impact on AI Deployment and Cost Efficiency
The reduction in costs for GPT‑6 Sol and Luna models significantly lowers barriers for integrating AI into business workflows, research, and customer service applications. By maintaining benchmark performance, OpenAI demonstrates that AI models can become more affordable without sacrificing quality, enabling wider adoption across industries. This shift could accelerate AI-driven automation, reduce operational expenses, and foster innovation in AI-powered solutions.
For companies and developers, the move means more budget-friendly options for deploying large language models at scale. It also signals a strategic focus on making AI accessible, not just advanced, which could influence competitors and the overall AI market landscape. The improved caching and inference efficiencies further support real-time applications and large-scale deployments, reinforcing the trend toward more cost-effective AI infrastructure.
As an affiliate, we earn on qualifying purchases.
Background on OpenAI’s Model Pricing and Performance
OpenAI’s GPT‑6 models, introduced in September 2026, marked a step forward in AI performance, with Astra representing the top-tier, high-cost models designed for maximum output quality. The new Sol and Luna models serve as mid-tier options aimed at balancing cost and capability, with prices cut by half compared to previous GPT‑5.6 models. Prior to this, OpenAI’s pricing structure involved higher operational costs, which limited large-scale deployment for many users.
The recent release emphasizes cost efficiency through technological improvements, including caching and inference optimizations, which allow the models to operate at lower expenses. Independent evaluations, such as those from Artificial Analysis, confirm that these models retain their performance levels, with scores comparable to or exceeding previous benchmarks in many areas. The focus on reducing hallucinations and improving refusal rates also reflects ongoing efforts to enhance reliability and safety in AI outputs.
cost-effective AI inference hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Remaining Questions About Model Capabilities and Limitations
While benchmark scores remain stable, it is not yet clear how these models perform in real-world, long-term deployments, especially in complex or specialized tasks. The noted regressions in knowledge tasks suggest potential limitations in detailed or comprehensive outputs, which could impact certain use cases. Additionally, the long-term effects of reduced presentation quality and shorter responses are still being evaluated by users and analysts.
OpenAI has not yet disclosed detailed operational metrics or user feedback from broader deployment, so the full impact of these cost reductions on performance in diverse applications remains to be seen.
AI caching and inference optimization devices
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in Evaluating and Deploying GPT‑6 Models
OpenAI is expected to continue monitoring the performance of GPT‑6 Sol and Luna in various applications, gathering user feedback and operational data. Future updates may include further tuning to address identified regressions, especially in knowledge-heavy tasks. Additionally, the company might expand caching and inference optimizations to enhance efficiency further.
Developers and organizations are advised to test these models within their workflows to assess suitability, especially in tasks requiring detailed output or comprehensive analysis. OpenAI may also release additional tools or guidelines to optimize model usage at the new price points.
affordable AI development platforms
As an affiliate, we earn on qualifying purchases.
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 is priced at roughly half the cost of GPT‑5.6 models, with input costs at $2.00 per 1 million tokens and output at $10.00. Luna costs about 50-60% less, with input at $0.10 and output at $0.50 per million tokens.
Do the benchmark scores indicate these models are less capable?
No, independent evaluations show that benchmark scores remain stable or improve slightly, indicating performance levels comparable to previous models despite lower costs.
Are there any trade-offs in model quality or reliability?
Some regressions in knowledge and presentation quality were observed, and the models tend to refuse to answer more often, which could impact certain detailed or complex tasks.
What does this mean for AI deployment in businesses?
The lower costs enable broader adoption of large language models in various workflows, potentially reducing operational expenses and increasing automation capabilities.
Will OpenAI release more updates or improvements soon?
OpenAI is likely to continue refining these models based on user feedback and performance data, possibly releasing updates to address current limitations.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
