GPT-5.6 Sol Shows a “50% Off” Price. What Does That Actually Mean?
GPT-5.6 Sol
An OpenAI AI model intended for complex professional tasks.
API
A way for software to use another service.
token
A small unit used to measure AI text.
What happened
OpenRouter, a service that lists ways to access AI models, shows a “50% off” label for OpenAI’s GPT-5.6 Sol. Its page lists prices of $2.50 per million input tokens and $15 per million output tokens.
That does not automatically mean OpenAI cut GPT-5.6 Sol’s official base price in half. OpenAI’s model documentation lists standard API prices of $5 for input and $30 for output per million tokens. OpenRouter’s own table shows those higher figures crossed out beside the lower ones.
Why there can be more than one price
A developer can use an AI model directly from its maker. They can also use it through a service that connects to several providers. OpenRouter is one such service. Its page lists routes through OpenAI, Azure, and Amazon Bedrock, with different prices and operating data.
A token is a small unit used to measure how much text an AI reads or produces. Input tokens are the material sent to the model. Output tokens are the material it returns. For projects that handle a lot of text, a change in either price can change the total bill.
Why this matters
The story is not simply that one AI became cheaper. It shows that the route to an AI can matter as much as the model name. If the displayed OpenRouter rate applies to a developer’s use, both the input and output prices are half of the standard rates shown by OpenAI.
But a lower rate can have limits. It may be a promotion, apply only through one route, or change over time. The available pages do not explain the offer’s end date or every condition. OpenAI also notes separate pricing rules for very long inputs.
What can be checked
OpenRouter currently displays a 50% discount and the $2.50/$15 rates for GPT-5.6 Sol. OpenAI’s official documentation lists $5/$30 as its standard text-token rates. OpenRouter also shows other providers with other prices. These facts support a narrower conclusion: pricing can differ by access route.
The Hacker News submission has 385 points and 219 comments. That measures attention from that community. It does not prove the discount terms or confirm that any price is correct.
What remains unclear
The public pages do not say how long the discount lasts, who can receive it, or whether it covers every kind of request. They also do not show an OpenAI announcement that changes Sol’s standard $5/$30 price.
The next useful checks are the terms attached to OpenRouter’s offer, future updates to OpenAI’s rate card, and the rate shown on an actual bill. For AI users, comparing costs means checking the model, the route, and the conditions together.
Why can GPT-5.6 Sol have two prices?
📰 Full story: GPT-5.6 Sol Shows a “50% Off” Price. What Does That Actually Mean?
The same AI can cost different amounts through different services.
OpenRouter
A service that offers access to several AI models.
GPT-5.6 Sol
An AI model made by OpenAI.
token
A unit used to count text handled by an AI.
💡 The gist
- OpenRouter shows GPT-5.6 Sol at 50% off.
- OpenAI lists a higher standard API price.
- The service route can change the final cost.
OpenRouter is a service for using many AI models.
It lists OpenAI’s GPT-5.6 Sol model.
Its page says “50% off.”
It shows $2.50 for one million input tokens.
It shows $15 for one million output tokens.
Tokens are small units for counting AI text.
Input is text sent to the AI.
Output is text returned by the AI.
OpenAI’s own model page lists different standard prices.
It lists $5 for input.
It lists $30 for output.
OpenRouter also shows those older amounts beside its lower rates.
So why can both pages exist?
People can use an AI directly from its maker.
They can also use it through another service.
OpenRouter is one of those services.
Its page lists several provider routes.
Those routes can have different prices.
That does not prove every user gets the lower price.
A discount may have rules.
It may last only for a limited time.
It may work only through one route.
The public page does not explain every rule.
OpenAI also has special pricing for very long inputs.
That is why teams should check more than a headline.
They should read the rate card.
They should check the route they will use.
They should check the bill after using it.
Hacker News users are watching this story.
The link has 385 points and 219 comments there.
Those numbers show attention.
They do not prove the price or the discount rules.
The main lesson is simple.
An AI name is not the whole price.
The path used to reach that AI matters too.
💬 A simple summary of the GPT-5.6 Sol price cut
The main takeaway is that Sol is cheaper, but only under certain conditions, and users report very different results. The numbers come from individual users.
- The 50% price is probably a temporary OpenRouter promotion. The native OpenAI API and Azure or Amazon routes are not necessarily discounted, and one report says only OpenAI-hosted Sol gets the OpenRouter discount. People guess it may attract customers or test demand, but that is unconfirmed.
- One user saw no clear usage jump on the discount day and reported about 101 billion tokens the day before. That is one person’s observation.
- Reports conflict: some users say Sol overthinks simple tasks, while another says a game reverse-engineering job fell from weeks to about 30 minutes. These are personal experiences.
- Grok 4.6 is also disputed: one user says it is about as capable and faster, while another says it is not equally strong. The thread does not prove a winner.
- Privacy matters too. A commenter says OpenRouter is not private inference; Tinfoil advertises hardware-based protection, but it is a different kind of service.
mature digest at 440 comments (revision 3). We fetched 300 comments and sampled 120 across the thread. These are HN users’ reports, not independently verified facts.
Two price tags for one AI
📰 Full story: GPT-5.6 Sol Shows a “50% Off” Price. What Does That Actually Mean?
GPT-5.6 Sol has a cheaper tag at one AI shop.
OpenRouter
A service that lets people use AI.
GPT-5.6 Sol
The name of an AI made by OpenAI.
OpenRouter is an AI shop.
It has GPT-5.6 Sol.
GPT-5.6 Sol is made by OpenAI.
OpenRouter shows a half-price tag.
Reading costs $2.50 there.
Writing costs $15 there.
OpenAI has a regular price tag too.
Reading costs $5 there.
Writing costs $30 there.
The OpenRouter tag is half.
Half of $5 is $2.50.
Half of $30 is $15.
The same AI can use different shops.
Different shops can charge differently.
We do not know when the sale ends.
Hacker News is talking about it too.
It has 385 points and 219 comments.
That means many people noticed.
It does not prove the price is right.
💬 GPT-5.6 Sol’s price cut, explained for a 5-year-old
Sol became cheaper in one place, but the answer depends on the shop and the job you give it.
- OpenRouter may be running a short sale. The OpenAI, Azure, and Amazon shops may not have the same sale, and even OpenRouter may discount only the OpenAI-run version. The reason may be to find more customers, but nobody knows for sure.
- One person counted about 101 billion tokens the day before the sale and did not see many more users on the sale day. That is only one person’s count.
- Some people say Sol thinks too long about easy jobs. Another says a hard game investigation went from weeks to about 30 minutes. Some people say Grok 4.6 is just as good and faster; others disagree.
- People who have secrets to protect also compare safety, not just price. OpenRouter and Tinfoil offer different kinds of protection, so the cheapest choice is not always the same choice.
mature digest at 440 comments (revision 3). We fetched 300 comments and sampled 120 across the thread. These are HN users’ reports, not independently verified facts.
💬 What the HN discussion says about GPT-5.6 Sol’s 50% OpenRouter cut
The discussion focuses on the discount’s scope, provider choice, inconsistent user experience, and privacy. Performance and usage figures are individual reports, not independent benchmarks.
mature digest at 440 comments (revision 3). We fetched 300 comments and sampled 120 across the thread. These are HN users’ reports, not independently verified facts.