The AI token price cut OpenAI is reportedly weighing will help you less than the headline promises. Your invoice is the price per token times the tokens your team burns, and agent tools grow the second number faster than any vendor can shrink the first. Budget by what a finished task costs.
Think of a lower price per litre. If the car you just bought drinks three times as much, the sign at the pump isn’t your problem.
Key points
- Is an AI token price cut really coming from OpenAI? No, not yet, as far as anyone can confirm. The Wall Street Journal reported on June 11 that OpenAI is considering major cuts to token prices and expects similar moves from Anthropic.
- Why wouldn’t a cheaper token mean a cheaper bill? The bill has two parts and the vendor controls only one.
- What does Uber’s budget tell a 50-person company? That usage isn’t value. Fortune reported on May 26 that Uber’s COO, Andrew Macdonald, finds it hard to connect Claude Code usage to useful consumer features.
- Which three numbers should you write down first? Put three numbers side by side before you decide whether a price cut matters to you.
Is an AI token price cut really coming from OpenAI?
No, not yet, as far as anyone can confirm. The Wall Street Journal reported on June 11 that OpenAI is considering major cuts to token prices and expects similar moves from Anthropic. OpenAI hasn’t published numbers, a start date or a list of models. Treat it as a reported plan, not a price you can budget against.
The motive is easy to see. Forbes reports that Sam Altman called token costs a huge issue for enterprise customers, and said it has become nearly a meme for companies to announce they’ve blown through their quarterly token budgets. A vendor hears that and sees churn risk. A buyer should hear a budgeting problem that the vendor’s pricing page won’t fix.
A caveat on sourcing. The Journal’s original sits behind a paywall, so we’re leaning on Forbes’s account of it.
Why wouldn’t a cheaper token mean a cheaper bill?
The bill has two parts and the vendor controls only one. A chat prompt uses a small, predictable number of tokens. A coding agent that reads a codebase, tries a fix, fails and tries again can use many times more per task. Cheaper units invite more units, and the invoice follows the second number. That’s our analysis, not a measurement.
Here’s a hypothetical, made up by us. A team spends $10,000 a month. A 30% price cut would bring the same work to $7,000. Now the team moves to agents that loop, and tokens per task double. The bill lands at $14,000 despite the cut. Price fell, spend rose. Nothing in that sum is exotic, which is the point.
Carry the same hypothetical one step further, still ours. Last month the team spent $10,000 and merged 200 pull requests, so a finished task cost $50. After the cut and the move to agents it spends $14,000 and merges 260, which sounds like progress. The cost per finished task is now about $54. Output rose 30%, spend rose 40%, and the price cut got the credit. Nobody sees that without counting the pull requests. Hardware prices follow the same rule, and robots vs human labour cost turns on usage rather than the sticker.
A sceptic will call a price cut plain good news, and for steady usage that’s right. If you pay by the token and your volume holds, a 30% cut is a 30% saving, and a price war between two big vendors tends to help buyers over a year or two. Take the saving. Just keep it out of the forecast until tokens per task have held still for a quarter, because teams adopting agents rarely hold still.
What does Uber’s budget tell a 50-person company?
That usage isn’t value. Fortune reported on May 26 that Uber’s COO, Andrew Macdonald, finds it hard to connect Claude Code usage to useful consumer features. “That link is not there yet,” he said. Fortune adds, citing other reports, that Uber used up its 2026 AI coding budget in four months, with an internal leaderboard ranking teams by use.
A 50-person firm can copy something better. Rank teams by finished work, not by tokens. Merged pull requests, closed tickets and approved drafts per team per month cost nothing to count, and they reward the team that gets the job done on fewer tokens. A leaderboard should make cheap, finished work look good.
Treat the budget detail as secondhand. Forbes attributes a similar claim to Uber’s CTO in April, while Fortune cites other reports, so the source of the four-month figure is unsettled. If it’s wrong, the arithmetic above still holds.
A leaderboard of usage buys usage, in our view. Uber has the cash to learn that in public. A firm with 50 staff and one finance lead learns it in a quarterly review. If your team celebrates tokens burned, you’re running a spending contest, and the vendor wins it.
Which three numbers should you write down first?
Put three numbers side by side before you decide whether a price cut matters to you. The first is what you spent last month. The second is how many finished tasks that spend produced, such as merged pull requests, closed tickets or approved drafts. The third is who can say stop. The first sits on the invoice.
The other two usually aren’t written anywhere.
Post them as one line in the finance channel every month. Three months of that line is a trend, and a trend is the only thing that makes a vendor conversation about price worth having.
Then settle three things. Work out what a finished task costs today and whether that has moved in three months. Name the single workflow that burns the most tokens, and decide what you’d stop doing if its budget halved. Confirm a spending cap exists, with a named owner.
If you can’t answer the first, a vendor’s price cut is noise. What AI actually costs a Canadian business covers the line items people forget, and the hidden time tax of AI work covers the cleanup hours that never show on a token invoice. Anthropic’s recent Opus 4.8 price hold is the same lesson from the other vendor, and Ramp’s AI Index shows who is being billed. A cheaper token buys nothing if the model makes things up, which is why Jev’s no-writing design is worth a look.
One practical step if a cut does arrive. Ask your vendor in writing if your contract passes a lower list price through or keeps you on the rate you signed. Committed-spend deals can lag the public price page, and the answer should be in your paperwork before it’s in your forecast.
Real figures from OpenAI would move us, as would per-task reporting and hard spend caps from vendors, since each would make price matter more than we’ve allowed. Until they arrive, pull last month’s invoice and count the finished tasks.
Frequently asked questions
Is OpenAI cutting its token prices?
Reportedly, yes. The Wall Street Journal says OpenAI is considering major cuts, but the company hasn’t announced figures, dates or models, so it isn’t a confirmed price change.
Will cheaper tokens lower my company’s AI bill?
Only if your usage stays the same. Agent and coding tools can use many times more tokens per task than chat, so a unit price cut can be outrun by volume.
How should a small business budget for AI usage?
Track cost per finished task, set a spending cap with a named owner, and review the biggest workflow each month. Token price alone tells you very little.
Archive entry dated June 12, 2026, written and fact-checked on October 8, 2026. Sources are linked on the claims they support.
Token prices only tell half the story, as our look at GPT-5.5 token efficiency showed.
