GPT‑6 Sol And Luna: OpenAI Cuts Prices In Half And Lets The Benchmarks Stay Flat
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: GPT‑6 Sol And Luna: OpenAI Cuts Prices In Half And Lets The Benchmarks Stay Flat on ThorstenMeyerAI.com

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

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

TL;DR

OpenAI has released GPT‑6 Sol and Luna models at 50% lower prices than GPT‑5.6, with performance benchmarks remaining steady. The move aims to expand AI accessibility by lowering operational costs without sacrificing quality.

OpenAI has introduced GPT‑6 Sol and GPT‑6 Luna models at prices that are approximately 50% lower than their GPT‑5.6 predecessors, without any decline in benchmark performance, marking a significant shift in AI cost structures.

The new models, GPT‑6 Sol and Luna, were launched on September 22, 2026, with pricing reductions based on improvements in caching and inference efficiency. GPT‑6 Sol now costs $2.00 per 1 million tokens for input and $10.00 for output, while Luna costs $0.10 and $0.50 respectively, compared to previous prices of double that. Despite the lower prices, independent evaluations from Artificial Analysis show that the models maintain similar or improved performance in key benchmarks, with GPT‑6 Sol scoring 48 on the AI Index and Luna scoring 37, well above median scores for their class. Cost savings are primarily achieved through technological enhancements, such as caching, which reduces the expense of serving these models. The models also show improvements in hallucination reduction, with Sol decreasing hallucination rates from 92% to 60%, and Luna from 93% to 77%, although with a trade-off in willingness to answer questions. Certain knowledge benchmarks, however, saw regressions, attributed to changes in presentation quality and output completeness. The models offer adjustable reasoning effort levels, allowing users to balance cost and performance, with Luna being notably cheaper at lower effort levels. OpenAI also enhanced prompt caching capabilities, enabling more efficient reuse of context and further reducing operational costs.
At a glance
updateWhen: announced September 22, 2026
The developmentOpenAI announced on September 22, 2026, that it is offering its GPT‑6 Sol and Luna models at half the previous prices, with unchanged benchmark scores, to improve cost-efficiency in AI deployment.

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.

Implications for Cost-Effective AI Deployment

This price reduction significantly lowers the financial barrier for deploying large language models in commercial applications, enabling broader adoption across industries. Maintaining benchmark performance ensures that quality remains consistent, while the improvements in hallucination reduction enhance reliability for customer-facing and research tasks. The move may pressure competitors to follow suit, accelerating the shift toward more affordable AI solutions. For businesses, these models offer a more viable path to integrating advanced AI capabilities without prohibitive costs, potentially transforming workflows and product offerings.
Amazon

AI language model API access

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background on OpenAI’s Model Pricing and Performance

Prior to this release, OpenAI’s GPT‑5.6 models were priced higher, with limited options for cost-efficient scaling. The company has historically balanced model performance with operational costs, but recent technological advancements in caching and inference have enabled significant price reductions. The launch of Astra, the top-tier model, underscored OpenAI’s focus on delivering the best results regardless of cost, but the new Sol and Luna models shift the focus toward accessibility and cost-efficiency. Independent evaluations, such as those from Artificial Analysis, have consistently shown that while higher-tier models outperform in some benchmarks, the gap is narrowing, and cost-efficiency is becoming a critical factor for widespread adoption.

Unanswered Questions About Long-Term Performance

It is not yet clear how these models will perform over extended periods or in diverse real-world applications, especially concerning their ability to maintain low hallucination rates and output quality in complex tasks. The impact of reduced presentation quality on knowledge work benchmarks also warrants further investigation.

Next Steps for Adoption and Evaluation

OpenAI is expected to continue refining these models, with further updates on performance, stability, and cost-efficiency. Industry adoption will likely increase as organizations test the models in production environments. Additional independent assessments and user feedback will shed light on their long-term viability and practical benefits, while OpenAI may expand its caching and inference optimizations to other models.

Key Questions

How much cheaper are GPT‑6 Sol and Luna compared to previous models?

GPT‑6 Sol costs roughly half of GPT‑5.6, at $2.00 per 1 million input tokens and $10.00 per 1 million output tokens. Luna costs about 50-60% less, at $0.10 and $0.50 respectively, representing approximately a 50% price reduction overall.

Do the new models perform as well as their predecessors?

Independent evaluations show that GPT‑6 Sol and Luna maintain or improve performance in key benchmarks, with Sol scoring 48 and Luna 37 on the AI Index, both above median scores for their class. However, some knowledge benchmarks saw regressions, and quality in detailed outputs may vary depending on tuning and use case.

What technological improvements enabled the price cuts?

OpenAI credits advances in caching and inference efficiency, including 90% discounts on cached input reads, for enabling lower operational costs. These improvements allow the models to be served at a fraction of previous costs without sacrificing performance.

Will these models replace higher-end options like Astra?

No. Astra remains the top-tier model for tasks requiring the highest quality, regardless of cost. Sol and Luna are positioned as more affordable options suitable for broader deployment where cost-efficiency is critical.

Are there any drawbacks to using the new models?

Some evaluations indicate regressions in presentation quality and completeness for knowledge work tasks, and a trade-off in hallucination reduction tactics may lead to more refusals rather than incorrect answers. Users should test models within their specific workflows before full adoption.

Source: ThorstenMeyerAI.com

FALL

Fall Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

The Orchestration Layer Arrives: What Anthropic’s Finance Agents Mean for Bloomberg, FactSet, and Wall Street

Anthropic releases new AI-powered financial agent templates and connectors, positioning Claude as an orchestration layer over major data providers, challenging Bloomberg’s dominance.

The Door: Why the Interface Is Worth More Than the Model

SpaceX’s $60 billion purchase of a coding interface highlights the growing importance of interface ownership over AI models in distribution and control.

Apple Wants Blacklisted Chinese RAM — And That Tells You How Bad The Squeeze Got

Apple is lobbying US authorities to buy Chinese-made memory chips from CXMT, raising concerns over supply chain and national security amid ongoing chip shortages.

APC–2 – A professional record cutter for producing original playback discs

The APC–2 is a newly introduced professional record cutter designed for producing high-fidelity playback discs in real time, available via select partners.