OpenAI has released new AI models, 'GPT-6 Sol' and 'GPT-6 Luna,' offering higher efficiency at lower costs.
Imagine this: You wake up in the morning and tell your smartphone AI assistant, “Organize my work schedule for today and help me fix the Python code I was working on yesterday.” In the past, whenever you assigned these tasks to an AI, you had to worry about cost and speed. You would mentally run a calculator, wondering, “How much will it cost to process this complex command?”
But now, you can breathe a little easier. On September 22, 2026, OpenAI released a new series of AI models, GPT-6 Sol and GPT-6 Luna, to kill two birds with one stone: “efficiency” and “accessibility.” Source 3, Source 5
Why is this important?
Until now, AI technology has focused all its capabilities on how much smarter it could become. Only perfect mastery of human language and complex reasoning were considered the highest values. However, smart AI consumes a massive amount of energy to calculate, and the costs were inevitably high.
For general users or businesses, there are many times when an “AI that is sufficiently smart and can be used without burden in daily life” is more desperately needed than an “extremely smart but very expensive AI.” Sol and Luna, announced this time, strike right at this point of “cost-effectiveness.” OpenAI explains that these models “make existing intelligence more efficient and accessible.” Source 5 In other words, we have entered an era where we can utilize cutting-edge AI technology for daily automation tasks or simple repetitive jobs at a lower cost.
Easy to understand: AI’s ‘Weight Class-Based Custom Services’
Simply put, this update is like systematically dividing the ‘weight classes’ of AI models. Think of it like a hospital.
If you need to undergo a complex and difficult brain surgery, you definitely need a skilled specialist (a top-tier model like GPT-6 Astra). But it’s similar to how you don’t need to find a university hospital professor just to get a cold medicine prescription or general health consultation. The key is to select and use the appropriate AI for the situation.
- GPT-6 Sol: A ‘smart practitioner.’ It is optimized for agent tasks (where AI uses tools on its own to achieve goals) that require self-judgment, such as programming and complex work automation. Source 9
- GPT-6 Luna: A ‘fast secretary.’ It is highly efficient for tasks where response speed is important and work is repetitive, such as chat and large-scale data classification. Source 6
Compared to the previous generation of GPT-5.6 models, these models provide the same performance using fewer tokens (the minimum unit for AI to read and understand text) for the same tasks. Source 7 Think of it as a car engine where the performance remains the same, but the fuel efficiency has dramatically improved, allowing you to go further.
Current situation: Practicality for everyone
Deployment of GPT-6 Sol and Luna has already begun. Source 3 Paid users can immediately utilize these models in ChatGPT Work or Codex, and a particularly encouraging fact is that GPT-6 Luna is also accessible to free users or lower-tier subscribers. Source 10
One change users will welcome most is the improvement regarding ‘hallucinations’ (the phenomenon where AI plausibly fabricates information that is not factual). Research results show that both Sol and Luna experience fewer hallucinations than previous models when performing at peak capacity. Source 4 This means AI is now providing more reliable information, which will further increase the reliability of daily work.
What will happen next?
In the future, the way we use AI will become more segmented based on individual preferences and purposes. You will call Sol when coding complex tasks and call Luna for light conversation or data organization.
Price competitiveness is also very strong. The cost has become very affordable, at $2 per 1 million input tokens for GPT-6 Sol and $0.10 for Luna. Source 3 This means more companies and developers can create various AI-based services with lower barriers to entry. In the future, smaller, faster, and cheaper AIs will be playing the role of your reliable assistants everywhere in your daily life.
MindTickleBytes AI Reporter’s View
In the competition of just pouring out smarter models, the paradigm is now shifting to an ‘efficiency competition’ where anyone can enjoy practical benefits. This is the clearest sign that technology is moving out of the lab and rooting itself deeply into our daily lives.
References
- OpenAI’s GPT-6 Sol doubles its accuracy rate - for half the cost - ZDNET
- GPT-6 Sol and Luna: pricing, benchmarks, choice · GPTunneL
-
[GPT-6 Sol and Luna push the cost efficiency frontier Artificial Analysis](https://artificialanalysis.ai/articles/gpt-6-sol-and-luna-push-the-cost-efficiency-frontier) -
[OpenAI launches GPT-6 Sol and Luna, boasting lower… TechCrunch](https://techcrunch.com/2026/09/22/openai-launches-gpt-6-sol-and-luna/) -
[GPT-6 Luna - API Pricing & Providers OpenRouter](https://openrouter.ai/openai/gpt-6-luna) - OpenAI upgrading ChatGPT and Codex with two more GPT-6 models
- OpenAI выпустила GPT-6 Sol и GPT-6 Luna: модели… / Хабр
- Anthropic and OpenAI Drop New High-Efficiency Models - CNET
- Complex programming
- High-volume data classification and chat
- High-difficulty academic writing
- Less hallucination than previous models
- More expensive than previous models
- Provided free to all users
- Simple alarm setting
- Programming, automation, and agent tasks
- Image generation only