🔍 Read the full analysis: What Does Cutting GPT‑6 Sol And Luna Prices In Half Mean For Benchmarking? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI has announced a 50% price reduction for GPT‑6 Sol and Luna models, significantly lowering costs for AI tasks. This development impacts benchmarking standards and the economics of AI deployment, with confirmed improvements in cost and hallucination rates but some regressions in knowledge tasks.
OpenAI has cut the prices of its GPT‑6 Sol and Luna models by 50%, making them the most affordable options in the company’s lineup. This price reduction, announced on September 22, 2026, is intended to expand the practical applications of AI by lowering operational costs, especially for tasks that previously might have been too expensive to automate. The move is significant because it shifts the economic landscape of AI benchmarking and deployment, affecting how models are evaluated and used across industries.
Both models, GPT‑6 Sol and Luna, now cost half their previous prices, with GPT‑6 Sol dropping from $4 to $2 per 1 million tokens for input and from $20 to $10 for output. GPT‑6 Luna’s costs are halved from $0.20 to $0.10 per 1 million tokens for input, and from $1.20 to $0.50 for output. According to OpenAI, these reductions are driven by improvements in caching and inference technology, passing savings directly to users.
Independent analysis by Artificial Analysis confirms that the cost per task has roughly halved, with the cost of running a typical task at max effort dropping from approximately $1.99 to $1.06 for Sol and from $0.20 to $0.07 for Luna. Despite the lower prices, the models maintain comparable or improved performance metrics in some areas, including hallucination reduction and coding capabilities, though some regressions in knowledge tasks have been observed.
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
How Price Cuts Alter AI Benchmarking and Usage
The 50% price reduction for GPT‑6 Sol and Luna models fundamentally changes the economics of AI deployment. Lower costs mean that tasks previously deemed too expensive can now be automated at scale, broadening the scope of AI applications across industries such as customer service, research, and content generation. This shift may lead to a reevaluation of benchmarking standards, as models become more accessible for large-scale testing and comparison. Furthermore, the improved cost-efficiency could accelerate innovation by enabling smaller organizations to deploy advanced AI solutions without prohibitive expenses.
However, the impact extends beyond economics. The analysis indicates that while hallucination rates have decreased—improving factual accuracy—some knowledge-based tasks have seen performance regressions, highlighting the importance of task-specific testing. Overall, the price cut emphasizes the trade-offs involved in optimizing AI models for cost versus quality, influencing future benchmarking and development priorities.
As an affiliate, we earn on qualifying purchases.
Background on AI Pricing and Benchmarking Trends
Prior to this announcement, OpenAI’s GPT‑5.6 models set the standard for high-performance language models, with pricing reflecting their advanced capabilities. The release of GPT‑6 Astra earlier this year marked a shift towards models emphasizing intelligence and capability, but at a higher cost. The recent price cuts for Sol and Luna represent a strategic move to democratize access to powerful AI, aligning with industry trends toward more affordable, scalable solutions.
Benchmarking has traditionally focused on performance metrics such as accuracy, hallucination rate, and reasoning ability, often at high costs. The introduction of more affordable models challenges these standards, as lower-cost models can now be evaluated more frequently and across broader datasets. This development may lead to a reevaluation of what constitutes state-of-the-art performance, especially as cost becomes a more central factor in model selection.
cost-efficient AI deployment tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Long-Term Effects
It remains unclear how these price reductions will influence the broader AI benchmarking landscape over the coming months. While initial data shows maintained or improved performance in some areas, the observed regressions in knowledge tasks suggest that further testing is needed to understand the models’ full capabilities. Additionally, how the market and competitors will respond with their own pricing strategies is still uncertain, as is the long-term impact on model development priorities.
As an affiliate, we earn on qualifying purchases.
Next Steps for Model Evaluation and Market Response
Expect ongoing benchmarking studies to assess the models’ performance across diverse tasks, especially in knowledge-intensive domains. OpenAI is likely to release further updates or tuning adjustments to address observed regressions. Industry analysts and competitors will monitor these developments closely, potentially prompting new pricing strategies and model improvements. Additionally, organizations considering adoption should conduct thorough testing to evaluate whether the lower-cost models meet their specific needs without sacrificing critical quality aspects.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much cheaper are GPT‑6 Sol and Luna models now?
GPT‑6 Sol’s costs are reduced from $4 to $2 per 1 million tokens for input and from $20 to $10 for output. Luna’s costs are halved from $0.20 to $0.10 per 1 million tokens for input, and from $1.20 to $0.50 for output.
Do the price cuts affect the models’ performance?
Initial analysis indicates that cost per task has roughly halved, with improvements in hallucination rates and coding performance. However, some knowledge and presentation quality metrics have shown regressions, so performance varies depending on the task.
What are the main benefits of these price reductions?
The reductions lower operational costs, enabling broader deployment and testing of AI models, and potentially accelerating innovation while making advanced AI more accessible to smaller organizations.
Will these changes influence AI benchmarking standards?
Yes, the affordability of these models is likely to lead to more extensive benchmarking, which could reshape what is considered state-of-the-art, especially as cost becomes a more prominent factor in model evaluation.
Are there any downsides to the new models?
Some observed regressions in knowledge-related tasks and presentation quality suggest that lower costs might come with trade-offs in certain areas, so thorough testing is recommended before deployment for specific use cases.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
