SHEYU.AI舍予基业
AI Model Watch

OpenAI cuts prices again: should bosses switch AI now or wait?

OpenAI's GPT-6.2 Astra and Sol cut input and output prices by 20%, but the real question for bosses is how much their teams can reliably produce each week.

·4 min read
OpenAI cuts prices again: should bosses switch AI now or wait?
AI-generated illustration, not a news photograph

OpenAI has released two new models, GPT-6.2 Astra and GPT-6.2 Sol, with input and output prices 20% below the previous generation and cached input down 60%. For companies using AI for content, customer service and daily office work, this is not spectator news: it changes the monthly tool bill and the timing of any switch. Our view is that the number worth calculating is not the unit price but how much a team can reliably produce every week.

What actually changed

According to IT Home on 18 November, OpenAI released two models with different positioning. GPT-6.2 Astra targets complex professional projects, with improvements in professional work, scientific research, software development and visual design. GPT-6.2 Sol targets daily work, balancing efficiency and general capability, and can complete tasks with fewer tokens.

On price, Astra's standard API rate is $8 per million tokens for input, $0.4 for cached input and $40 for output — 20% lower than GPT-6 Astra on input and output, and 60% lower on cached input. Sol keeps the previous generation's pricing: $2 input, $0.1 cached input and $10 output.

One detail on availability is worth noting. IT Home reports that from 17 November the two models are open to Plus, Pro, Enterprise and Edu users of ChatGPT Work and Codex, and developers can call them through the API. Eligible Enterprise and Edu users still need a workspace administrator to enable them. In other words, released does not mean you can open it and use it; corporate accounts still have to go through an administrator switch.

Why the big labs keep cutting prices

Price cuts are not charity; they are forced.

According to an analysis by Origami, around July 2026 xAI, OpenAI, Meta and Moonshot made moves within two weeks, and with several comparable-quality providers present at once, prices naturally drift down. The article also cites a structural reason: almost every model family now has an "economy tier" that handles most daily tasks at a fraction of the flagship price, and open-source models that can be deployed in-house put a ceiling on closed API pricing.

What does this mean for operators? Three things.

First, AI features you cut a year ago because they were "too expensive" deserve a fresh calculation at today's prices. Call summaries, automatic replies, document information extraction — scenarios that did not add up then may be worthwhile now.

Second, do not throw every task at the most expensive model. Give simple, repetitive work to the economy tier and keep the hard work for the flagship; that is the most direct way to save money.

Third, the point Origami's analysis makes most sharply: a low token price does not mean "almost free". Actual cost is decided by usage — a million small requests cost far more than a thousand large ones — and low prices invite overuse, so the bill climbs before you notice. The costs of systems engineering, monitoring and failed retries never appear on the price list.

What bosses should really weigh

If you stop to recalculate every time a model cuts its price once a month, you will get nothing else done.

Shu Yi, executive director of Sheyu Jiye, put it plainly: customer acquisition through content is not about one viral piece, but about whether you can produce steadily every week and be seen consistently.

The same holds for choosing AI tools. Cheaper models solve only the problem of being able to afford it, not the problem of producing something. Many bosses buy accounts and memberships, and still the team cannot produce three decent pieces of content a week — not because the model is weak, but because nobody connects topic selection, production and publishing into one line.

So the judgment to make now is not "wait for the next cheaper model" but: is this thing I am doing repeated every day, and going wrong every day? Shu Yi makes another point: in digitalisation, start with the thing that repeats every day and goes wrong every day. The price cut only lowers the cost of trial and error on that thing; acting is still up to you.

One approach you can try now

If what you are building is the content customer-acquisition line, SHEYU AGENT is one option to put on the shortlist.

According to its official website, it brings together 16 industry advisers and 34 zero-threshold tools, covering Xiaohongshu image-and-text posts, AI posters and copywriting, with one login working across desktop, mobile and web. Its official positioning is "the boss's first AI team" — meaning the first AI team a boss owns, not a ranking.

What it can do is take over the content production stage: generating images, writing copy, producing drafts by topic, and tracking whether major AI systems mention your brand. What it cannot do is make judgments for you. Shu Yi has said that AI writes the first draft, and the boss's judgment and real experience are the most valuable part of the content. Feed it brand facts and real cases and the output sounds like you; do not, and what comes out is correct nonsense.

One more thing needs saying: the money saved by cheaper models does not automatically become your profit. Cheaper tools test whether you have a steady rhythm of topics and publishing.

Takeaway: recalculate the AI features you shelved as too expensive, but judge any switch by whether it turns into steady weekly output — not by the unit price on the rate card.

閱讀繁體中文版 →