GPT-5.6 Sol at 50% Off on AI Gateways Is Real — But Three Things the Announcements Do Not Say Out Loud (2026 Verified)

A reader sent us a screenshot: “GPT-5.6 Sol, 50% off — is this real?”

Short answer: yes, and it is not just one provider. As of our verification today (2026-08-26), both Cloudflare AI Gateway and Vercel AI Gateway are running the same promotion, ending on the same date: September 18, 2026.

But the interesting part of this story is not the words “50% off.” It is the three things we hit while verifying it — one of which is that two official pages disagree on what the original price is.

1. Verifying it: both official changelogs, quoted

Cloudflare AI Gateway (posted 2026-08-19)

We pulled the official changelog page directly. Key sentences, unedited:

“GPT-5.6 Sol is available through AI Gateway, and for a limited time you can use it at 50% off. If you are already using AI Gateway, point to the openai/gpt-5.6-sol model and the discounted pricing applies automatically — no promo code needed.”

“The promotion is available for Unified Billing users only (not Bring Your Own Keys).”

“The promotion runs through September 18, 2026.”

The pricing table on that page, as we saw it:

Vercel AI Gateway

The Vercel changelog goes further and lays out all three service tiers — and notes that OpenAI cut the list price mid-promotion:

• List before the cut: Default US$2.50 / US$15, Flex US$1.25 / US$7.50, Priority US$5 / US$30
• List after the OpenAI cut: Default US$4 / US$20, Flex US$2 / US$10, Priority US$8 / US$40
• With the 50% gateway discount applied: Default US$2 / US$10, Flex US$1 / US$5, Priority US$4 / US$20

(All per million tokens, for requests up to 272K context.)

From the Vercel page:

“BYOK requests bill at your own rate with OpenAI.”

“Cached tokens, cache writes, long-context requests above 272K, and the US regional rates all move by the same proportion.”

Both official sources confirm it. The promotion is real. The screenshot our reader sent (US$2 / US$10 / US$0.25, against US$4 / US$20 / US$0.50) matches the “post-list-cut, then halved” version of the numbers.

2. What we actually caught: the same “50% off” has two different baselines

This is the finding worth writing down.

At our verification timestamp, the Cloudflare page still listed the standard price as US$5 / US$30, while the Vercel page states the list has already dropped to US$4 / US$20.

Both are official. Neither has a typo. The difference is the baseline date: OpenAI cut the GPT-5.6 Sol list price on 2026-08-21 (we verified that separately in GPT-5.6 Sol 20% cut is promotional pricing, guaranteed only through 11/21), and gateway announcement pages do not necessarily go back and resync older figures.

So:

• Read the Cloudflare table, and you conclude half price is US$2.50 / US$15
• Read the Vercel table, and you conclude half price is US$2 / US$10

That is a 20% gap. And note the subtlety: the cheaper figure is not cheaper because Vercel discounts harder. It is cheaper because its baseline is newer.

We are not going to declare which page is stale — we record what we saw at verification time. But here is the usable rule:

During any promotion, ignore the percentage, read the unit price you actually pay. The denominator behind a “% off” can move. The unit price cannot lie.

3. Three conditions the announcements do not say loudly

3.1 BYOK gets nothing

Both providers say the same thing: the discount is for Unified Billing users only. If you route through the gateway with your own OpenAI key, you bill at your own OpenAI rate — full price.

This matters more than it sounds. Plenty of teams adopt a gateway precisely so they can keep their own key and use the gateway only for caching, rate limiting, and logs. That setup gets zero benefit here. To capture the discount you must load credits with the gateway provider — moving the billing relationship from the model vendor to an intermediary.

Whether that trade is worth it depends on how much you trust that intermediary, not on the 50%.

3.2 After 9/18, nothing is promised

Both deadlines are 2026-09-18. Neither changelog says what the price reverts to, or whether the promo will be extended.

That mirrors the list price itself: the OpenAI “20% lower” list is guaranteed only “at least through November 21, 2026.” Which means the US$2 / US$10 you see today is a promotion stacked on top of a promotion — with two separate expiry dates, 9/18 and 11/21.

Building an annual budget on that number means modelling twelve months on two coupons that both expire. Same structural trap we took apart in The annual-plan trap behind AI price cuts.

3.3 Only standard-sized requests get cheap

The Vercel line “all move by the same proportion” sounds generous, but read it in reverse: long-context requests above 272K, cache writes, and Priority (fast mode) all scale up proportionally too. Half off does not make the surcharges disappear.

Priority at half price is US$4 / US$20 — exactly the Default tier list price. If fast mode is your default habit, this promotion means “you now pay what everyone else pays at full price.”

4. Our take: why the discount landed on these platforms

To be clear about what follows: this is a third-party firm critique, not an established fact. Neither OpenAI nor the platforms have confirmed any intent.

Research firm SemiAnalysis raised a point about this promotion: OpenRouter and Vercel AI Gateway account for a negligible share of actual OpenAI usage, yet they are among the few public data sources the industry uses to estimate model market share — their analytical influence far exceeding their real volume.

Their argument, in substance: if a half-price promotion more than doubles Sol token volume on those two platforms within a month, investors might read that as a decisive win over competitors — and that reading may not reflect the real market.

Our view: the claim is unverifiable, but it points at something genuinely useful for readers —

Public leaderboards measure the volume that runs on that leaderboard, not the volume in the world. That number can be bought with a single promotion. Your procurement decision should not rest on someone else chart.

The three questions that actually matter have nothing to do with who is ranked first:

1. Is the output quality good enough on your task? (Someone else benchmark is not your workload.)
2. Can you absorb the post-promotion price? (If not, do not migrate — migration costs more than the savings.)
3. Are you willing to move the billing relationship from the model vendor to an intermediary for a discount? (That is an architecture decision, not a pricing one.)

5. Bottom line

The news is real and the numbers are real — but “50% off” is the least informative part of it.

What actually determines your bill is in the fine print: whether you are on Unified Billing, whether your requests are standard-sized, and what you plan to do after 9/18.

Just testing the model for a few days? Take the discount, it is a good deal. Choosing a model for a product that has to run for a year? Make the decision on the post-9/18 price, not the price today.

Official sources

• Cloudflare AI Gateway changelog: https://developers.cloudflare.com/changelog/post/2026-08-19-gpt-5-6-sol-discount/
• Vercel changelog (three-tier pricing after the list cut): https://vercel.com/changelog/gpt-5-6-sol-is-now-50-percent-off-a-lower-price
• Vercel changelog (original promo announcement): https://vercel.com/changelog/gpt-5-6-sol-is-50-off-on-ai-gateway-for-the-next-month

Related verification on this site

• Tool page: ChatGPT / OpenAI plans and free tier
• GPT-5.6 Sol 20% cut is promotional pricing
• The annual-plan trap behind AI price cuts
• Claude Sonnet 5 API price lock, verified
• DeepSeek API price increase, verified

All prices quoted from official changelog pages, verified 2026-08-26. Promotional terms can change at any time — check the official page before committing. The SemiAnalysis critique is a third-party opinion, unconfirmed by OpenAI or the platforms; we report it as information, not as our position.