GPT-5.6 Sol Is $4/$20, Not $5/$30, and That Price Is Promotional Through at Least November 21
- What happened
- OpenAI's live API pricing docs put GPT-5.6 Sol at $4.00 input and $20.00 output per 1M tokens on short context, and state that the promotional pricing is available "at least through November 21, 2026" with no published end date and no published post-promotional rate.
- Why it matters
- The number most teams are budgeting against is explicitly temporary, and OpenAI has published no rate for what follows it. Our own directory entry also carried an incorrect $5/$30, which matched no row on the live card and has since been corrected.
- What to do
- Re-price your Sol workloads at $4/$20 today, then model a second scenario for after November 21, 2026 and decide in advance which workloads you would move. Unpin from the gpt-daybreak-blue-latest and gpt-daybreak-red-latest aliases, which re-point on OpenAI's schedule.
The verdict
Budget GPT-5.6 Sol at $4.00 in and $20.00 out per 1M tokens on short context. Then budget a second time for the day that number stops being true.
OpenAI's pricing docs carry a line next to Sol that changes how you should read the whole rate card: "GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026." That is a floor, not a ceiling. There is no published end date, no published post-promotional rate, and no commitment beyond the floor.
Sol is still the most cost-effective frontier model on the board, and our directory rating does not move on this. What moves is your planning horizon.
The same rate card lists GPT-6 Astra at $10.00/$50.00 on short context, and shows that OpenAI's two Daybreak aliases currently resolve to Sol and GPT-5.6-Cyber. Both details are below.
What happened
Nothing was announced. OpenAI's published rate card settles a price conflict we flagged and left open earlier this week, and it resolved in a direction we did not expect.
The live rates
Per 1M tokens, gpt-5.6-sol:
| Mode | Short context | Long context |
|---|---|---|
| Standard | $4.00 in / $20.00 out | $8.00 / $30.00 |
| Cached input | $0.40 | $0.80 |
| Cache writes | $5.00 | $10.00 |
| Batch | $2.00 / $10.00 | $4.00 / $15.00 |
| Flex | $2.00 / $10.00 | $4.00 / $15.00 |
| Fast | $8.00 / $40.00 | $16.00 / $60.00 |
Cached input at $0.40 is a tenth of the standard input rate. If your workload has a stable system prompt or a repeated document context, that line is worth more to your bill than the headline number is.
The correction we owe you
Our directory entry for GPT-5.6 Sol stored $5 input and $30 output. That figure matches no row on the live card. It is not the short-context rate ($4/$20) and it is not the long-context rate ($8/$30). It was our error, not a vendor change, and the entry has since been corrected to $4/$20 short context and $8/$30 long.
We are saying that in the open because a directory that quietly fixes its own numbers is not a directory anyone should trust. If you priced a Sol migration off our page in the last few weeks, re-run it: your input cost is 20% lower than we told you, and your output cost is 33% lower on short context.
What else was on the same page
Three things worth recording from the same rate card, because they are cheap to check now and expensive to discover later.
GPT-6 Astra holds at $10.00/$50.00 short context and $20.00/$75.00 long, unchanged from the figures in our launch coverage. Fast mode remains unavailable for Astra under EU data residency, per OpenAI's launch material.
gpt-daybreak-blue-latest and gpt-daybreak-red-latest are aliases, not models. Today they point at gpt-5.6-sol and gpt-5.6-cyber respectively. They will re-point as OpenAI ships new Daybreak models, which means anyone pinned to an alias has agreed in advance to a model swap they will not be notified about.
Why it matters
A promotional price with a published floor and no published ceiling is a different financial instrument from a list price, and it should be modelled differently.
The gap between $4/$20 and Sol's nearest non-promotional comparison is the size of the exposure. If Sol reverts toward the $5/$30 tier that GPT-5.5 still carries, a workload sized at today's rate gets 25% more expensive on input and 50% more expensive on output overnight. That is not a prediction. OpenAI has published no post-promotional rate at all, so any number you assume is your own invention. The honest planning assumption is that the rate is unknown after November 21 and that you will get little warning.
The second-order problem is worse for teams that have already committed. Frontier model pricing has moved down repeatedly this year, which trains everyone to treat the direction of travel as permanent. A promotion with a stated floor date is OpenAI telling you, in writing, that this particular decline has an expiry attached.
What changes for you
Re-price now, at the correct rate, and put a date in the calendar.
If you are running Sol today, the news is good and immediate: your real cost is lower than our page said. Recalculate against $4/$20, and check whether cached input at $0.40 applies to your traffic shape before you optimize anything else.
If you are planning a Sol commitment that runs past November, do not build the model on $4/$20 alone. Build it twice, once at the promotional rate and once at a rate you would still accept, and know in advance which of your workloads you would move if the price changed. The models most exposed are the high-output ones, because the output side of the card is where the headroom sits.
And if you are pinned to a Daybreak alias, unpin. An alias that re-points on OpenAI's schedule is a pricing and capability change you have pre-authorized.
FAQ
What is the promotional price, exactly?
$4.00 input and $20.00 output per 1M tokens on short context, $8.00/$30.00 on long context. Cached input is $0.40 short context and $0.80 long. Batch and Flex both halve the standard rate.
What happens on November 21, 2026?
Unknown, and that is the point. OpenAI publishes the date as a floor ("at least through November 21, 2026"), not an end date. There is no published expiry and no published post-promotional rate, so any number you plug in after that date is your own assumption, not OpenAI's.
Was the $5/$30 figure a price cut, then?
No. $5/$30 was our directory error. It matched no row on the live card, neither the short-context rate nor the long-context one. Nothing changed on OpenAI's side between our entry and this pull.
Should I pin to gpt-daybreak-blue-latest or gpt-daybreak-red-latest?
No. Those are aliases, not models. Today they point at gpt-5.6-sol and gpt-5.6-cyber, and they re-point as OpenAI ships new Daybreak models. Pinning to one is pre-authorizing a model swap, with its own pricing, that you will not be notified about.
What to do
- 1 Re-price every GPT-5.6 Sol workload at the correct rate: $4.00 input and $20.00 output per 1M tokens short context, $8.00/$30.00 long context. If you priced off our directory entry, your real cost is 20% lower on input and 33% lower on output than we told you.
- 2 Check whether cached input at $0.40 per 1M applies to your traffic. A stable system prompt or repeated document context makes that line worth more to your bill than the headline rate.
- 3 Put November 21, 2026 in the calendar as a pricing review date. OpenAI publishes it as a floor, not an end date, and has published no post-promotional rate at all.
- 4 Model any commitment that runs past November twice: once at $4/$20 and once at a rate you would still accept. Identify the high-output workloads you would move first, because output is where the exposure sits.
- 5 Unpin from gpt-daybreak-blue-latest and gpt-daybreak-red-latest. They are aliases currently pointing at gpt-5.6-sol and gpt-5.6-cyber, and they will re-point as OpenAI ships new Daybreak models.
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.