On September 22, 2026, something unusual happened. Anthropic released a new AI model called Claude Opus 5.5. About ninety minutes later, OpenAI released two new models of its own, GPT-6 Sol and GPT-6 Luna. Neither company waited for the other to make the first move, and both ended up cutting their prices at the same time. People are calling it an AI price war, and if you use AI tools for work, it genuinely changes what you should be paying. Here is what actually happened, and more importantly, which one makes sense for you.
Quick Answer
- Claude Opus 5.5: $4 per million input tokens, $20 per million output tokens, Anthropic’s strongest model for demanding work
- GPT-6 Sol: $2 per million input tokens, $10 per million output tokens, half the price of Opus 5.5
- GPT-6 Luna: $0.10 per million input tokens, $0.50 per million output tokens, built for high-volume, simpler tasks
- For the best possible quality, Opus 5.5 currently comes out ahead, even accounting for its higher price
- For the best value on everyday tasks, GPT-6 Sol is hard to beat
- For huge volumes of simple work, Luna is dramatically cheaper than anything else on the market right now
What Actually Happened
Anthropic released Claude Opus 5.5 as an upgrade to its most powerful model, priced 40% lower than the previous version while performing better on coding and computer-use tasks. OpenAI, releasing GPT-6 Sol and GPT-6 Luna just an hour and a half later, cut its own prices by half compared to its earlier promotional rates. This came only nineteen days after OpenAI had launched its top-tier model, GPT-6 Astra, so the company was already mid-rollout when it decided to also undercut Anthropic on price.
Neither company directly compared its new release to the other’s in its own marketing. Anthropic measured Opus 5.5 against OpenAI’s older models. OpenAI measured Sol and Luna against its own previous generation. It took independent testing labs, like Artificial Analysis, running the exact same prompts through every model, to get an honest side-by-side answer.
What Each Model Actually Costs
Here is the full picture, in plain numbers, per million tokens (tokens are roughly chunks of words, so a million tokens is a lot of text):
| Model | Input Price | Output Price |
|---|---|---|
| Claude Opus 5.5 | $4 | $20 |
| GPT-6 Sol | $2 | $10 |
| GPT-6 Luna | $0.10 | $0.50 |
A few things worth knowing that do not show up in the headline numbers:
- Both Opus 5.5 and Sol charge the same $0.20 per million tokens for cached content, which is content the AI has already seen before in your conversation. For people running AI agents that reread the same instructions repeatedly, this cached rate ends up mattering more than the headline price.
- OpenAI charges extra once a single request goes over 272,000 input tokens, roughly double the input price and one and a half times the output price. Anthropic does not do this, it charges the same rate across its entire context window. If you regularly work with very long documents or conversations, this difference can add up.
Which One Actually Performs Better
Price alone does not tell the whole story, so here is what independent benchmark testing found.
On a well-known coding test, Terminal-Bench, Opus 5.5 and OpenAI’s flagship GPT-6 Astra both landed at the exact same score, 59.6%, at their strongest settings.
When Artificial Analysis ran GPT-6 Sol and Claude Opus 5.5 through the same set of tests at every effort level, a clear pattern showed up: Sol is the cheaper way to get a decent result, right up until you need a really strong one. Past a certain quality level, only Opus 5.5 could reach it at all, and interestingly, it did so more cost-effectively than Sol’s best possible attempt. In other words, pushing Sol to its absolute limit to match Opus 5.5’s normal performance ends up costing more than just using Opus 5.5 in the first place.
Which One Should You Actually Use
This is the part that matters most, so here is a simple way to decide based on what you actually do.
Pick Claude Opus 5.5 if:
You are working on something that genuinely needs to be right, like production code, detailed analysis, or complex reasoning where mistakes are costly. It costs more per task, but for hard problems, it currently delivers the best results for the price, not just the best results.
Pick GPT-6 Sol if:
You want strong, reliable results for everyday work, like drafting content, answering questions, building typical software features, or running AI agents that need to be good but not necessarily the best in the world. It gives you most of the top-tier quality at half the cost.
Pick GPT-6 Luna if:
You are running something at large scale, like classifying thousands of messages, summarizing short pieces of content, or powering a simple chatbot that handles a high volume of straightforward requests. At a fraction of a cent per task in many cases, Luna makes tasks financially viable that would be too expensive to run through a bigger model at scale.
A practical rule of thumb: start with the cheapest model that can actually do your task well, and only move up if it is not cutting it. Testing a few real examples of your own work through each option before committing is worth the extra hour it takes, since benchmark scores do not always match how a model performs on your specific use case.
Where to Access These Models
GPT-6 Sol and Luna are currently available through OpenAI’s API, and through ChatGPT for Plus, Pro, Business, Enterprise, and Edu account holders. Free account users only get access to Luna, and only through the desktop app for now. Claude Opus 5.5 is available through Anthropic’s API and Claude’s paid plans.
Frequently Asked Questions
What is the difference between Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna?
Opus 5.5 is Anthropic’s top-tier model built for demanding tasks. Sol is OpenAI’s mid-tier model, offering strong results at half the price. Luna is OpenAI’s budget model, built for simple tasks at very high volume.
Which AI model is the cheapest?
GPT-6 Luna is by far the cheapest, at $0.10 per million input tokens and $0.50 per million output tokens, which is a small fraction of what the other models cost.
Which AI model gives the best results?
For the hardest tasks, Claude Opus 5.5 currently delivers the strongest results, and independent testing shows it actually becomes more cost-effective than cheaper models once you need that level of quality.
Why did Anthropic and OpenAI cut prices on the same day?
Both companies released new models on September 22, 2026, within about ninety minutes of each other, both at significantly lower prices than their previous offerings. Neither has confirmed the timing was intentional, but it has been widely described as a price war.
Can I use these models for free?
GPT-6 Luna is available for free users, but only through the ChatGPT desktop app. Sol and Opus 5.5 currently require a paid plan or API access.
Is a more expensive AI model always better?
Not necessarily for every task. For simple, high-volume work, a cheaper model like Luna often performs perfectly well and saves significant cost. The more expensive models are worth it specifically when task difficulty is high enough that cheaper models start struggling.
