OpenAI Slashes Costs For GPT‑6 Sol And Luna Without Changing Benchmark Outcomes
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: OpenAI Slashes Costs For GPT‑6 Sol And Luna Without Changing Benchmark Outcomes on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get smart everyday buys 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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

OpenAI has slashed the prices of its GPT‑6 Sol and Luna models by 50%, with no significant change in benchmark performance. The move aims to make AI more accessible without sacrificing quality, affecting how businesses deploy AI solutions.

OpenAI has cut the prices of its GPT‑6 Sol and Luna models by 50%, without any significant change in their benchmark scores. This move, announced on September 22, 2026, aims to make advanced AI models more affordable for a broad range of applications, from research to enterprise workflows. The price reduction is achieved through improvements in caching and inference efficiencies, passing savings directly to users.

OpenAI introduced GPT‑6 Sol and Luna on September 22, 2026, as part of its Astra family, with prices now at half those of GPT‑5.6 models. GPT‑6 Sol now costs $2.00 per 1 million tokens for input and $10.00 for output, down from $4 and $20 respectively, representing a 50% reduction. GPT‑6 Luna is priced at $0.10 per 1 million input tokens and $0.50 per output token, halving previous costs of $0.20 and $1.20. Despite the lower prices, independent analysis from Artificial Analysis indicates that the models’ benchmark scores, including intelligence and coding evaluations, remain roughly consistent with prior models, with some improvements and regressions noted.

OpenAI attributes the cost savings to enhanced caching and inference techniques, including a 90% discount on cached input reads. The models also exhibit improved hallucination rates, with Sol reducing hallucinations from 92% to 60%, and Luna from 93% to 77%, partly due to increased refusal rates. However, some knowledge-work benchmarks show regressions, particularly in presentation quality and completeness of deliverables, which may impact specific workflows.

At a glance
updateWhen: announced September 22, 2026
The developmentOpenAI announced on September 22, 2026, that it has reduced the costs of its GPT‑6 Sol and Luna models by half, while maintaining comparable benchmark scores, emphasizing cost efficiency.

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.

GPT‑6 Sol
$4 / $20 → $2 / $10
GPT‑6 Luna
$0.20 / $1.20 → $0.10 / $0.50

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.

GPT‑5.6 Sol
$1.99
GPT‑6 Sol
$1.06
GPT‑5.6 Luna
$0.18
GPT‑6 Luna
$0.07

The effort dial moves cost more than the model choice

Model and effortIntelligence IndexCost 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

Already on GPT‑5.6 Sol or Luna? The move is mostly a price cut. Re‑test first if your output is a document someone reads, not data a system consumes.
Shelved an automation on cost? Token prices halved and the effort dial adds another order of magnitude. Re‑run the business case.
Choosing between labs? The question is no longer which model is smartest, but which clears your quality bar at the lowest cost per task.
ThorstenMeyerAI.comSources: OpenAI (pricing, vendor benchmarks) and Artificial Analysis (independent evaluation and model pages). Figures as of 23 September 2026.

Impact on AI Deployment and Cost Efficiency

The price reductions for GPT‑6 Sol and Luna represent a significant shift in AI economics, lowering the barrier for businesses and developers to incorporate advanced language models into their products and workflows. By maintaining benchmark performance while halving costs, OpenAI enables more extensive automation, research, and customer-facing applications without increasing budgets. This move could accelerate AI adoption across industries, especially for tasks where cost constraints previously limited usage. Additionally, the improvements in hallucination and refusal rates suggest better safety and reliability, further broadening potential use cases.

Amazon

AI model cost optimization tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

OpenAI’s Model Pricing and Performance Evolution

OpenAI’s recent model releases have focused on balancing performance with cost. The GPT‑6 Astra models, introduced in September 2026, marked a new generation with larger context windows and improved inference. The company’s strategy involves optimizing hardware and software efficiencies, such as caching, to reduce costs. Prior to this, GPT‑5.6 models were the standard, with higher prices that limited widespread deployment. The announcement aligns with industry trends toward more affordable AI models that do not compromise on capability, following similar moves by competitors like Anthropic.

Independent evaluations, such as those from Artificial Analysis, have shown that while costs have decreased, the models’ benchmark scores remain competitive, with some metrics even improving. However, some knowledge-based evaluations have experienced regressions, indicating that the tuning for cost efficiency may impact certain aspects of model performance, especially in delivering comprehensive, high-quality outputs.

Amazon

AI inference efficiency hardware

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 still unclear how these models will perform in large-scale, real-world deployments over time, especially regarding their ability to handle complex, multi-turn interactions and produce high-quality, comprehensive outputs consistently. Additionally, the long-term impact of the tuning adjustments on accuracy and factuality remains to be fully understood, as some knowledge benchmarks showed regressions.

Amazon

AI development caching solutions

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Upcoming Developments and Monitoring

OpenAI is expected to continue refining its models and infrastructure, with upcoming updates likely to focus on balancing cost efficiency with performance. Users and developers should monitor official communications for further improvements, especially in areas such as hallucination reduction, output quality, and multi-task performance. Additionally, further independent evaluations will clarify how these models perform across diverse applications and industries.

Amazon

cost-effective AI tokenizers

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

How much are GPT‑6 Sol and Luna models now costing?

GPT‑6 Sol now costs $2.00 per 1 million input tokens and $10.00 per 1 million output tokens, while GPT‑6 Luna costs $0.10 and $0.50 per 1 million tokens for input and output, respectively.

Do the models’ benchmark scores indicate they are less capable?

No. Independent analysis shows that benchmark scores remain stable or improved slightly in some areas, despite some regressions in knowledge and presentation tasks.

What technical improvements enabled these cost reductions?

OpenAI improved caching and inference efficiencies, including a 90% discount on cached input reads, which significantly lowered operational costs.

Are there any trade-offs in quality with the new models?

Some regressions in knowledge task performance and deliverable presentation have been observed, which may impact workflows requiring detailed and comprehensive outputs.

What should users consider before switching to these models?

Users should test the models in their specific workflows, especially if their tasks rely heavily on detailed, high-quality outputs, as some presentation and knowledge metrics have declined slightly.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
FALL

Fall Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

Interim Results Show SenseTime-W’s AI Business Achieves New Milestones

SenseTime-W reports RMB 607 million profit and 28.2% growth in generative AI revenue in its latest interim results, signaling a strategic shift.

Revolutionizing AI Development: Meta’s Muse Spark 1.2 Is Here

Meta releases Muse Spark 1.2 and Muse Code, a new AI coding model and agent with co-training, long-horizon capabilities, and improved safety features.

Beyond Models: Tackling The Plumbing Challenges Of AI At Scale

Experts highlight that integration and orchestration bottlenecks, not model capabilities, are the main hurdles for large-scale AI deployment in 2026.

ShinyHunters · The New APT Model.

Analysis of ShinyHunters’ evolving operational model, AI-enabled tactics, and implications for enterprise cybersecurity post-2026.