Anthropic charges double and keeps winning. Now OpenAI might blink on price.
Both labs filed to go public within a week of each other. Days later, OpenAI was reported to be weighing token price cuts to slow Anthropic down. Strip out the IPO theater and one question is left for anyone paying an API bill: if the two biggest model labs start fighting on price, what lands in your invoice?
Photo by Piotr Makowski on Unsplash
The quick read
- Two IPO filings a week apart, then a report that OpenAI may cut token prices to chase Anthropic. Nothing has actually changed on either rate card yet.
- The twist worth your attention: Anthropic is the pricier lab and still gaining share, so the company under pressure to cut is the cheaper one. We work out what a cut would do to your bill, and the 50% discount you can already buy today.
Anthropic filed confidentially for an IPO on June 1, putting a $965B valuation on the table. OpenAI filed its own confidential paperwork seven days later, carrying a $852B mark from its March funding round. Then, on June 10, the Wall Street Journal reported that OpenAI was discussing cuts to what it charges, and that it would drop token prices if Anthropic moved first. OpenAI's line, paraphrased through the reporting: nothing decided, talks in flux.
That sequence is strange on its face. The cheaper lab is the one rattled enough to consider cutting further. To see why, you have to look at who is actually winning, and then at the rate cards, because the gap between the two does not point the way you would guess.
The lab that charges more is the one pulling ahead
Menlo Ventures' enterprise survey from late 2025 put Anthropic at 40% of enterprise LLM API spend, OpenAI at 27%, and Google at 21%. Three years ago OpenAI held roughly half that market alone. On revenue run-rate, Anthropic disclosed a $47B figure alongside its May funding round; OpenAI sits around $24B to $25B. One caution on those numbers: a run-rate annualizes a recent strong month, so it flatters both labs and is not the same as booked annual revenue. Anthropic's actual 2025 revenue was closer to $10B.
Even with that asterisk, the direction is clear. Anthropic is taking enterprise share and growing faster, and it is doing it while charging a premium. That is the part that should reframe how you read a price war: OpenAI is not the expensive incumbent trying to hold a line. It is the cheaper challenger watching a pricier rival eat its lunch, which is exactly the spot where cutting further starts to look tempting.
What the frontier costs today
Here are the current standard list prices, per million tokens, with a monthly bill for a mid-sized production app pushing 30M input and 6M output tokens. These are the numbers a price war would move.
| Model | Input / 1M | Output / 1M | 30M in / 6M out |
|---|---|---|---|
| Claude Fable 5 | $10.00 | $50.00 | $600 |
| OpenAI GPT-5.5 | $5.00 | $30.00 | $330 |
| Claude Opus 4.8 | $5.00 | $25.00 | $300 |
| Gemini 3.1 Pro | $2.00 | $12.00 | $132 |
| Gemini 3.5 Flash | $1.50 | $9.00 | $99 |
Read that against the market-share story and it gets odd. GPT-5.5 already undercuts Anthropic's flagship by half and ties Opus 4.8 on input while charging a bit more on output. Anthropic is the pricey option here, especially with Fable 5 at $600 a month for the same work that costs $330 on GPT-5.5. And Anthropic is still gaining. That is the wall OpenAI keeps running into: on this kind of workload it is not losing on price, so a price cut is a bet that price was the problem in the first place.
If the cut lands, here is the size of it
No percentage was ever reported, so anything specific is a guess. But it is a useful guess, because OpenAI's own discount tiers tell you the room it has to move. The table below takes that same 30M / 6M month on GPT-5.5 and models two cuts against where Opus 4.8 sits today at $300.
| GPT-5.5 scenario | Input / Output | Monthly bill | vs Opus 4.8 ($300) |
|---|---|---|---|
| Today, standard | $5.00 / $30.00 | $330 | +$30 |
| Hypothetical 25% cut | $3.75 / $22.50 | $247.50 | -$52.50 |
| Hypothetical 50% cut | $2.50 / $15.00 | $165 | -$135 |
A quarter off would flip GPT-5.5 from $30 above Opus 4.8 to about $52 below it. Half off would put a frontier OpenAI model at $165, nearly half of Opus and a bit over a quarter of Fable 5. For a team running serious volume, that is the difference between a few hundred and a few thousand dollars a month. Plug your real split into the cost calculator to see where your own bill would land at each step.
You can already buy the 50% cut
Here is the thing the WSJ headline buries: that 50% scenario is not hypothetical for work that can wait. GPT-5.5 batch and flex processing already runs at $2.50 input and $15 output, the exact halved rate, today, no announcement required. Cached input drops further to $0.50 per million. If a chunk of your traffic is overnight jobs, evals, or anything that tolerates latency, you are leaving the price war's prize on the table by paying standard rates right now.
And if the model does not have to be frontier-class, the cut is even steeper outside OpenAI entirely. Gemini 3.5 Flash handles a huge share of production traffic at $1.50/$9, which is $99 for that same month against GPT-5.5's $330. The price war people are waiting on between two labs has a third party already running it. Watch the GPT-5.5 long-context surcharge, though: prompts over 272K tokens bill at 2x input and 1.5x output, and that surcharge applies even on batch and flex.
What to actually do about it
Do not rearchitect around a rumor. Nothing about OpenAI's rate card has changed, and a confidential IPO filing means the pressure to look profitable could just as easily push prices the other way. The move is to capture the savings that already exist: route latency-tolerant work to batch or flex, turn on prompt caching, and push anything that does not need a flagship down to a Flash-tier model.
If a cut does land, it changes which frontier model is cheapest, not whether you should have been measuring. Keep your real input-output split in front of you so that the day a rate card moves, you can tell in a minute whether switching is worth the migration, instead of taking a headline's word for it.
Sources
- OpenAI weighs price cuts - CNBC on the WSJ report (no percentage given, talks in flux)
- OpenAI confidential IPO filing - CNBC (June 8, $852B last private valuation)
- Anthropic confidential IPO filing - Fortune (June 1, $965B valuation)
- Enterprise market share - Menlo Ventures (Anthropic 40%, OpenAI 27%, Google 21%)
- GPT-5.5 pricing - OpenAI ($5/$30 standard, $2.50/$15 batch, $0.50 cached input)
- Claude pricing - Anthropic (Fable 5 $10/$50, Opus 4.8 $5/$25)