IPO track ends the discount era
A hypothesis on one completed case, published as such. Zhipu AI removed first-purchase discounts 34 days after its January 2026 Hong Kong IPO and raised its GLM Coding Plan list 80% at 195 days. The proposed enabler is margin disclosure — a listed lab has to defend gross margin quarterly. Three adjacent cases do not support it, and the corpus-wide version of the idea has already failed.
What's happening — and why
What's happening: one lab went public and then systematically withdrew promotional pricing. Zhipu AI listed in Hong Kong on 2026-01-08, removed first-purchase discounts 34 days later, and by day 195 had relisted its GLM Coding Plan at $18 / $72 / $160 per month — roughly 26% higher effective at the entry tier and 40% at the top.
The proposed mechanism is margin disclosure. A listed company defends gross margin on a quarterly cadence, which makes a first-purchase promo expensive in a way a private lab does not feel, because the promo shows up as revenue quality rather than as growth.
Why this page exists at n=1: the corpus publishes hypotheses with their evidence base stated, and this one's base is a single completed arc. Three adjacent cases actively fail to support it. Moonshot AI ran both of its price moves before its listing process surfaced. 01.AI moved new revenue off its published card without touching the card. Anthropic widened its funnel 13 days after IPO-preparation reporting — the opposite of the predicted direction.
And the general version of this idea has already been tested and failed: funding events do not predict repricing in this corpus. Only the narrower listing sub-case retains any support at all.
How it works
Evidence over time
13 supporting · 6 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Zhipu AI | Aug 2024 | Land-grab stage: the GLM-4-Flash API is made free to the public during the China model price war, establishing free-and-cheap as the distribution strategy. |
| Zhipu AI | Sep 2025 | GLM Coding Plan launches with a $3 first month for Lite, pitched at roughly 3x Claude Pro usage for half the price — an explicit shot at Claude Code and Cursor economics. Standing price settles near $10/mo Lite, $30 Pro, $80 Max (billed quarterly at ~$30/$90/$240). |
| Zhipu AI | Jan 2026 | Listing signal: Zhipu (02513.HK) debuts on the Hong Kong Stock Exchange, the first foundation-model lab to go public anywhere, raising ~US$560M at a ~US$6.7B valuation with the retail tranche oversubscribed ~1,159x. |
| Zhipu AI | Feb 2026 | All GLM Coding Plan first-purchase discounts removed (ending the $3/$15 promo); standing prices settle at ~$10/$30/$80 per month. 34 days after listing — the tightest dated interval between a capital-markets event and a discount withdrawal in the corpus. |
| Zhipu AI | Jul 2026 | List reprice at +195 days: Coding Plan goes to $18/$72/$160 per month from roughly $10/$30/$80 (+80% Lite, +140% Pro, +100% Max on the monthly list). A -10%/-20%/-30% monthly/quarterly/yearly ladder is published with renewal rates printed, so the cheapest effective entry is $12.60/mo (+26%). A premium GLM-5 API line lands at $1.00-$1.40 in / $3.20-$4.40 out, roughly double the GLM-4.x flagships holding at $0.60/$2.20. |
| Moonshot AI | Jul 2025 | Land-grab stage: the Kimi K2 open-weight launch (1T parameters, 32B active) at roughly $0.15/M cache-hit input and $2.50/M output becomes the fastest-downloaded model on Hugging Face a day after release and is credited with igniting the China price war. |
| Moonshot AI | Apr 2026 | INVERTED ORDER. Kimi K2.6 reaches GA with the first material increase: input +58% ($0.60 to $0.95/M), output to $4.00/M, cache-hit input $0.16/M. Read contemporaneously (KuCoin, OpenRouter, accessed 2026-06-11) as commercialization proof ahead of a reported Hong Kong listing — 88 days BEFORE the formal process was reported. |
| Moonshot AI | Jul 2026 | INVERTED ORDER. Kimi K3 opens a new ceiling at $3.00 in / $15.00 out per 1M (cache hit $0.30, 1,048,576-token context) — 3.2x input and 3.75x output on K2.6, which is left unchanged, so this is a new rung rather than a repricing. Ships WITH a launch top-up rebate (10-30% of a single top-up as vouchers, 2026-07-15 to 2026-08-12) that no longer appears at its own URL as of 2026-07-30. |
| Moonshot AI | Jul 2026 | Listing signal arrives LAST: Yahoo Finance and Investing.com report a plan to list in Hong Kong within six months of the K3 debut, and The Information (2026-07-20) reports Moonshot seeking investor approval to begin the IPO process — 3 days after the K3 ceiling and 88 days after the K2.6 rise. |
| 01.AI | Jul 2026 | BYPASS VARIANT. Tech in Asia and finance.biggo.com report 01.AI targeting a 2027 Hong Kong IPO and repositioning as 'China's Palantir' — enterprise and government deployment over open-model distribution. |
| Moonshot AI | Aug 2026 | The window's one supportive datum, and still in the inverted order. Moonshot pulled Kimi Enterprise's only public price — $599/yr/seat with a disclosed feature list (enterprise data privacy, 2x monthly Agent credits, early access to Agent Swarm and Kimi Claw, Kimi Code 5x credits, dedicated support), live through July 30 — from the China-site kimi.com/zh-cn/business page, leaving a bare [email protected] contact and no rate. It lands in the same month Moonshot is reported to be seeking investor approval to begin a Hong Kong IPO process, making it a third pre-listing monetization tightening after the 2026-04-22 K2.6 input rise (+58%) and the 2026-07-16 K3 ceiling. The consumer Kimi ladder ($0 to $199/mo) and the per-token developer API are unchanged and fully public, so this is one track moving upmarket rather than a repricing. |
| 01.AI | Aug 2026 | THE BYPASS VARIANT COMPLETES — and it ends in deletion, not a reprice. Fourteen days after the reported 2027 Hong Kong IPO target and twelve days after TrueNorth shipped quoted-only, 01.AI posted a dated wind-down notice on platform.lingyiwanwu.com: new registrations and top-ups closed immediately, model trial and API calls stop 2026-09-03 at 24:00, and balance-refund applications are accepted through 2026-12-03 at 24:00, with a dedicated refund button and a [email protected] contact. 01.AI frames it as a deeper focus on enterprise AI solutions. The published rate card never moved — yi-lightning still lists at ¥0.99 per million tokens and yi-vision-v2 at ¥6, unchanged since October 2024 and re-verified on 2026-07-22 — so this vendor neither withdrew a discount nor raised a list price on its way toward a listing. It closed the shop with the prices still in the window. |
| 01.AI | Jul 2026 | BYPASS VARIANT, two days later: TrueNorth (万策) ships as Boss AI / Investor AI / TopSales AI plus forward-deployed engineers and sovereign deployments, sold entirely via 'Book a Consultation'. The ¥0.99/¥6 per-1M-token card is re-verified UNCHANGED the same day — no discount is withdrawn and no list price rises; new revenue simply moves off the published card. The 2-day ordering is our inference, not a company statement. |
Counterexamples
- Cerebras · Aug 2026 — A completed IPO with zero pricing consequence — the trend's most direct test in the corpus so far, and it failed. Cerebras Systems, the wafer-scale AI inference cloud positioned as an Nvidia challenger, completed its long-anticipated IPO in the week of 2026-08-03, pricing above its initial range and reportedly ~20x oversubscribed on debut before shares fell more than 58% from their first-day high inside the same week (Stocktwits, The Motley Fool, TradingKey, renaissancecapital.com). No pricing change accompanied the listing: the per-token inference API (GPT-OSS-120B, Gemma), the $5 free-trial credits, the $10 self-serve Developer tier and the fixed-price Cerebras Code subscriptions were all unchanged at capture. Note the ordering if anything: Cerebras converted its open Free tier into a one-time $5 credit trial on 2026-07-21, three weeks BEFORE the listing, not after it.
- Yellow.ai · Aug 2026 — A second listing event in the same fortnight with no repricing. Yellow.ai agreed to go public via a SPAC merger with Bluerock Acquisition Corp (NASDAQ: BLRK) at roughly $550M — a $300M stock deal plus a $5M equity facility and a $50M PIPE note — confirmed by an official PR Newswire release and reported consistently by CMSWire, Stock Titan, TradingView, Investing.com and Unite.AI, several framing the strategy as using the listing to roll up BPO and outsourcing firms into AI-native operations. Its free tier, $0.99 per-resolution overage and sales-led Enterprise motion were unchanged. A public-listing and capital-structure event that touched nothing a buyer pays.
- DeepSeek · Aug 2026 — The mechanism attacked directly, from the opposite side to MiniMax. DeepSeek published the largest price increase in the corpus this window — peak/off-peak API billing effective 16:00 UTC on 2026-08-16, peak hours 01:00-04:00 and 06:00-10:00 UTC, all other hours off-peak at exactly half the peak rate. V4-Flash cache-miss input goes from a flat $0.14 per 1M to $0.22 off-peak and $0.44 peak; V4-Pro from $0.435 to $0.66 and $1.32; the steepest relative jump lands on cache-hit input, DeepSeek's most aggressively marketed rate. Roughly 1.5-2.5x off-peak and 3-12x at peak. DeepSeek is privately held with no listing, no filing and no reported IPO process. If margin disclosure at listing were what ends a Chinese lab's discount era, the lab that ended its discount era hardest in this window did it with no capital-markets event at all.
- Anthropic · Jul 2026 — Widened the funnel instead of narrowing it. CNBC reported (2026-07-15) bankers lining up investor meetings for a large IPO, with 3Q26 profit above $1B; 13 days later the Claude.ai Team minimum fell from 5 seats to 2 ('For teams of 2 to 150') with every seat price held — Standard $20/$25, Premium $100/$125. A packaging move that makes the product cheaper to start, in the direction opposite to this trend.
- MiniMax · Jul 2026 — Listed in Hong Kong in January 2026 (~US$619M raised at ~$6.5B, +43% on debut) on roughly $53M of revenue against a ~$512M loss, and was still cutting six months later: MCP API-vlm cut from $0.06 to $0.01 per call effective 2026-07-22 and the Token Plan revamped to roughly double included tokens at unchanged $20/$50/$120 prices. A listed lab pricing at the floor is the direct refutation of the margin-disclosure mechanism.
- Together AI · Jul 2026 — Raised $800M at an $8.3B valuation on 2026-07-01 with the CEO declining to rule out an IPO the following year (Axios, 2026-07-02), then ran two further cut releases (2026-07-14 dedicated-inference restructure with cuts, 2026-07-21 Kokoro-82M TTS $10.00 to $4.00 per 1M characters) before its first increase 28 days later — reserved H100 $3.59 to $3.69 for 7-30 days. An IPO signal without a filing did not stop the cuts.
Trivia
-
Zhipu AI's follow-on capital did not buy buyers a lower price. It raised about $4B in a Hong Kong share sale on 2026-07-08, after a reported ~1,500% run-up in the stock, and 14 days later — on 2026-07-22 — raised the GLM Coding Plan instead of cutting it. Zhipu is also the corpus's only pricing arc dated to the day on both ends: discounts gone at +34 days after the 2026-01-08 listing, list price up 80% at +195 days.
-
Moonshot AI ended its discount era and launched a new discount in the same release. The 2026-07-16 Kimi K3 launch opened a $3.00/$15.00 per-1M ceiling — 3.2x the input and 3.75x the output of K2.6 — and shipped alongside a top-up rebate returning 10-30% of a single top-up as vouchers, advertised 2026-07-15 to 2026-08-12. By 2026-07-30 that promotion URL served a generic Help Center billing FAQ instead: the promo appears withdrawn ahead of its own stated end date.
-
Zhipu's 2026-02-11 "discount withdrawal" did not remove the discount — the 2026-07-22 card published it instead, as a standing -10% monthly / -20% quarterly / -30% yearly ladder with the renewal rate printed on each tier. So the cheapest possible entry moved from a $3 first month (2025-09) to a $12.60/mo annual floor: +26% against the prior effective rate while the monthly list rose +80% ($10 to $18). The steepest effective rise was the middle tier, Pro, at +68% ($30 to $50.40).
For buyers
Do not plan around this. With one completed case and three that fail to fit, a listing is not a reliable predictor of repricing, and treating it as one will produce more false alarms than saved budget. What is worth doing is cheap and general: if a vendor you depend on lists, re-read your renewal terms and check whether your discount is contractual or promotional. That is good practice regardless of whether this hypothesis survives, and it is the only action the evidence actually supports.
For vendors
The interesting question this raises is not whether to reprice after listing but whether your promotional structure survives quarterly scrutiny at all. A first-purchase discount that is invisible in a private company's metrics becomes a line item in a listed company's revenue-quality discussion. If you are on a listing track, the useful preparation is to know which of your discounts are structural (volume, commitment) and which are acquisition promos, because the second category is what comes under pressure first — and withdrawing it abruptly is what generates the churn.
Outlook — what to watch
This needs more completed cases before it means anything, and the corpus will get them: several labs are on listing tracks. The honest test is prospective — log the promotional structure before a listing and re-check at 30, 90 and 180 days. If two more labs run Zhipu's arc, this becomes a pattern; if the next one runs Anthropic's, it should be superseded. Note that the corpus-wide version of this idea — that funding events predict repricing — has already been tested and failed, so only the narrow listing sub-case remains live.
Bottom line
One lab removed discounts after listing and raised list prices 26% to 40% within 195 days. Three adjacent cases do not fit, and the general funding-events-predict-repricing version already failed. This is a hypothesis published with its n stated, not a pattern to plan around.
FAQ
Do AI companies raise prices after going public?
There is one completed case in the corpus. Zhipu AI removed first-purchase discounts 34 days after its 2026-01-08 Hong Kong IPO and relisted its GLM Coding Plan at $18 / $72 / $160 by day 195, roughly 26% to 40% higher effective. Three adjacent cases do not support the pattern, so it should not be generalised.
Why publish a trend based on one case?
Because the alternative is publishing it as a pattern later without the caveat. The corpus states evidence bases explicitly, and this one is n=1 complete with three non-fitting adjacent cases and a failed corpus-wide version. Readers can weigh it accordingly.
What is the proposed mechanism?
Margin disclosure. A listed company defends gross margin quarterly, which makes a first-purchase promotion expensive in a way a private lab does not feel — it shows up as revenue quality rather than growth. That is a plausible mechanism, but plausibility is not evidence, and Anthropic widened its funnel 13 days after IPO-preparation reporting.
Should I renegotiate before a vendor lists?
The evidence does not support urgency, but it does support hygiene: check whether your discount is contractual or promotional. Promotional discounts are the category that comes under pressure first in any margin-defence exercise, listing or not.