🔍 Read the full analysis: GPT‑6 Sol And Luna Now Half Price—OpenAI Keeps Benchmark Scores Flat on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
OpenAI announced that GPT‑6 Sol and Luna models are now priced at half their previous costs, while benchmark scores remain stable. This shift makes AI more accessible for broader applications, though some evaluation metrics show mixed results.
OpenAI has announced a significant price reduction for its GPT‑6 Sol and GPT‑6 Luna models, cutting costs by 50% compared to their GPT‑5.6 predecessors. The models now cost $2.00 per 1 million tokens for input and $10.00 for output for Sol, and $0.10 per 1 million tokens input and $0.50 for output for Luna, with the company citing improvements in caching and inference efficiency as the key to lower costs. Despite the price cuts, OpenAI reports that benchmark scores remain roughly flat, emphasizing cost efficiency over performance gains.
On September 22, 2026, OpenAI launched GPT‑6 Sol and Luna at half the previous prices, aiming to make advanced AI more affordable for a wider range of users and applications. The price reductions are attributed to improvements in caching techniques and inference processes, which allow the models to be served at lower costs, with savings passed directly to customers. For example, GPT‑6 Sol’s cost per task, evaluated by independent analysis firm Artificial Analysis, dropped from $1.99 to $1.06, while Luna’s cost decreased from roughly $0.18 to $0.07 per task.
Despite the lower prices, benchmark evaluations show that the models’ performance remains stable. Sol scores 48 on the Artificial Analysis Intelligence Index, well above the median of 25 for similar models, with Luna scoring 37 against a median of 12. However, some evaluation metrics, such as knowledge-work tasks, experienced regressions, with Sol dropping about 100 Elo points on GDPval‑AA v2.1, attributed to reduced presentation quality and omitted details. OpenAI states that the models now refuse to answer more often, which reduces hallucinations but can also lower accuracy and completeness.
OpenAI emphasizes that these models are designed for cost-sensitive applications, where trade-offs in completeness and presentation are acceptable. The company also highlights new caching features, including a dashboard and diagnostics tool, to help developers optimize their use of the models.
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 Accessibility and Cost Efficiency
The price cuts for GPT‑6 Sol and Luna significantly lower the barrier to entry for deploying advanced language models in commercial and research settings. By maintaining benchmark performance while halving costs, OpenAI enables smaller companies, startups, and research institutions to incorporate AI into their workflows without prohibitive expense. This shift could accelerate AI adoption across industries, from customer service to content creation, and foster innovation driven by more affordable AI tools.
However, the evaluation results suggest that cost savings may come with some trade-offs in model quality, particularly in knowledge accuracy and presentation. For users relying on detailed, well-structured outputs, testing remains essential to ensure suitability. Overall, the move underscores a strategic focus on democratizing AI access rather than pushing performance boundaries further.
As an affiliate, we earn on qualifying purchases.
Recent Developments in AI Pricing and Performance
OpenAI’s release of GPT‑6 Astra earlier this month marked a milestone in AI capability, but the focus was on pushing the frontier of intelligence. The subsequent introduction of GPT‑6 Sol and Luna shifts the emphasis toward cost efficiency, reflecting a broader industry trend of making AI more affordable. Prior to this, models like Anthropic’s Claude Opus 5.5 had also announced price cuts, but OpenAI’s approach combines significant price reductions with stable benchmark scores, setting it apart.
The evaluation by Artificial Analysis, published the same day, provides an independent assessment of the models’ performance and costs, confirming that the reductions are primarily cost-driven with only modest changes in intelligence scores. The models’ ability to deliver comparable results at half the previous price point marks a notable development in the AI market, particularly for applications where cost constraints are critical.
While the models’ performance remains stable overall, some specific knowledge tasks experienced regressions, attributed to tuning adjustments aimed at reducing hallucinations and improving safety. OpenAI has indicated that future updates may focus on balancing performance with cost and safety considerations further.
cost-effective AI development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Remaining Questions About Model Performance and Use Cases
It is not yet clear how the models will perform in real-world, long-term deployments, especially in tasks requiring high accuracy and detailed outputs. The reported regressions in some knowledge work benchmarks suggest that the models may be less suitable for tasks demanding comprehensive presentation and detailed deliverables. Additionally, the impact of increased refusal rates on user experience and workflow efficiency remains to be seen. The full extent of performance trade-offs and how they will influence different use cases is still developing.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Evaluation
Organizations interested in these models should conduct their own testing to verify suitability for their specific tasks, particularly in knowledge-intensive workflows. OpenAI is expected to release further updates, possibly refining the models’ presentation quality and balancing hallucination reduction with output completeness. Market adoption will likely accelerate as more users experiment with the models in various applications, and additional independent evaluations will clarify their long-term performance and value.
AI diagnostics and optimization tools
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 now?
GPT‑6 Sol’s cost per 1 million tokens has been reduced from $4 to $2 for input and from $20 to $10 for output. Luna’s costs dropped from $0.20 to $0.10 per input and from $1.20 to $0.50 per output, representing roughly 50% savings across the board.
Does the price cut affect the models’ performance?
No, according to OpenAI and independent evaluations, benchmark scores and capabilities remain roughly the same, although some knowledge-task performance has seen minor regressions.
Are there any trade-offs with the new models?
Yes, the models now tend to refuse answering more often, which reduces hallucinations but can also lead to less detailed or complete outputs in some cases.
What should users consider before switching to these models?
Users should test the models in their specific workflows to ensure performance meets their needs, especially for tasks requiring detailed, well-structured responses.
What is the significance of these price changes?
The reductions significantly lower the cost barrier for deploying advanced AI, potentially enabling broader adoption across industries and smaller organizations.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
