🔍 Read the full analysis: OpenAI Halves Prices For GPT‑6 Sol And Luna While Benchmark Results Stay Flat on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- 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 its GPT‑6 Sol and Luna models, effective September 22, 2026. Despite the lower costs, independent evaluations show that model performance remains largely unchanged, prompting analysis of the implications for AI deployment and quality.
OpenAI has announced a 50% reduction in the prices for its GPT‑6 Sol and Luna models, effective immediately from September 22, 2026. The move aims to make advanced AI more accessible for a broader range of applications, even as independent benchmark tests show no significant change in performance compared to previous models. This development is notable because it shifts the focus from improving AI capabilities to reducing operational costs, which could influence how businesses and developers adopt these models.
On September 22, 2026, OpenAI launched GPT‑6 Sol and Luna at prices half of those for GPT‑5.6, citing improvements in caching and inference efficiencies as the basis for cost reduction. The new pricing sets GPT‑6 Sol at $2.00 per 1 million tokens for input and $10.00 for output, down from $4 and $20 respectively. GPT‑6 Luna now costs $0.10 for input and $0.50 for output, compared to previous prices of $0.20 and $1.20. Despite these cuts, independent analysis by Artificial Analysis on the same day indicates that the models’ benchmark scores remain roughly on par with older versions, with some evaluations showing slight regressions.
Artificial Analysis’s evaluation reports that GPT‑6 Sol at maximum effort scores 48 on its Intelligence Index, well above the median of 25 for comparable models, with GPT‑6 Luna scoring 37 against a median of 12. The cost per task has halved, with GPT‑6 Sol costing approximately $1.06 per task and Luna around $0.07, despite both models using slightly more output tokens per task. Notably, the models show significant improvements in hallucination reduction, with Sol decreasing hallucination rates from 92% to 60%, and Luna from 93% to 77%. However, some knowledge benchmarks, such as GDPval‑AA v2.1, saw declines in performance scores, indicating potential regressions in certain knowledge-related tasks.
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 of Lower Costs Without Performance Gains
The price cuts for GPT‑6 Sol and Luna could make AI deployment more economically feasible for a wider array of applications, especially where cost constraints previously limited adoption. However, the lack of measurable performance improvements in benchmark tests raises questions about whether the cost savings come at the expense of quality in complex tasks. For businesses, this means that while operational costs may decrease, careful testing is necessary to ensure that model quality remains sufficient for specific use cases, especially those requiring high accuracy or detailed outputs.
This development could shift market dynamics, pressuring competitors to also lower prices or improve efficiency, and prompting users to reevaluate the trade-offs between cost and performance. It also underscores the importance of understanding the nuances of AI model capabilities, particularly in applications sensitive to hallucinations and factual accuracy.
As an affiliate, we earn on qualifying purchases.
Background on OpenAI’s Pricing and Performance Trends
OpenAI’s recent model releases have focused on balancing performance with operational efficiency. The launch of GPT‑6 Astra earlier this month marked a significant step forward in AI capabilities, but the subsequent price reductions for GPT‑6 Sol and Luna highlight a strategic shift toward making powerful models more affordable. Prior to this, GPT‑5.6 models were priced higher, with incremental improvements in performance and efficiency. The new models’ pricing reductions are attributed to advancements in caching and inference techniques, which reduce the cost of serving these models.
Independent benchmark evaluations, like those from Artificial Analysis, have historically tracked the performance of these models across various tasks, providing a baseline to assess the impact of cost-cutting measures. The recent results show that, despite the lower prices, the models’ scores in intelligence, coding, and hallucination metrics remain stable or slightly regressed, suggesting that OpenAI’s focus on cost efficiency is not necessarily translating into performance gains at this stage.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Model Quality and Use Cases
It remains unclear whether the performance metrics used in benchmarks fully capture the models’ effectiveness in real-world applications. The observed regressions in some knowledge tasks suggest that lower prices might come with trade-offs in the depth or presentation of outputs. Additionally, the long-term impact of these cost reductions on AI quality, user trust, and competitive positioning is still uncertain. Further testing across diverse tasks and use cases is needed to understand whether the models’ practical performance aligns with their benchmark scores.
As an affiliate, we earn on qualifying purchases.
Next Steps for OpenAI and Model Adoption
OpenAI is likely to continue refining its models and caching strategies, aiming to further reduce costs without sacrificing quality. Users and developers should monitor ongoing performance evaluations and conduct their own testing before fully integrating these models into critical workflows. Market competition may also drive other providers to pursue similar cost reductions or performance enhancements. Future updates from OpenAI could clarify whether these price cuts are sustainable and whether subsequent models will deliver measurable performance improvements alongside lower costs.
cost-effective AI inference hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why did OpenAI reduce the prices of GPT‑6 Sol and Luna?
OpenAI attributed the price reductions to improvements in caching and inference efficiencies, which lowered operational costs and allowed the company to pass savings to users without changing the models’ core capabilities.
Do the lower prices mean the models are less capable?
Not necessarily. Independent benchmark tests show that performance remains roughly the same, although some specific knowledge tasks have experienced slight regressions. The models’ core abilities appear unaffected by the price cuts.
Will the performance regressions affect real-world applications?
It depends on the use case. For tasks requiring detailed, well-presented outputs, regressions in presentation quality might matter. For simpler or pipeline-based tasks, the impact may be minimal.
How might this affect the AI market overall?
The price cuts could pressure competitors to lower their prices or improve efficiencies, potentially leading to broader adoption of advanced AI at lower costs, though the long-term effects on quality are still uncertain.
What should users do before adopting these models?
Users should test the models within their specific workflows to ensure that performance and quality meet their requirements, especially for tasks where accuracy and presentation are critical.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
