ZHIPU and MINIMAX HIP-3 perps: what sits underneath and the risks
Quote801 words
Breaks down the two HIP-3 equity perps on HK-listed Zhipu and MiniMax: GLM-5.2 and M2.7 benchmarks, valuation against US labs, and risks such as sub-6% float and cornerstone lock-up expiry around July 8.
Key claims
Zhipu ~$120-150B at over 1,000x sales; MiniMax ~$20B at ~250x
GLM-5.2 scores 62.1 on SWE-bench Pro vs GPT-5.5 at 58.6
Cornerstone lock-ups expire around July 8
$ZHIPU and $MINIMAX HIP-3 markets (two of the hottest open-weight AI labs btw) are live on Hyperliquid.
Trading at a fraction of the US frontier labs, this could be a generational entry. Yet, it's a risky bet.
I aped both, and here is why:
First, what these markets actually are:
Both are HIP-3 equity perps launched by @tradexyz referencing the Hong Kong listings (Zhipu is HK2513, MiniMax is HK0100).
Until today, exposure to either meant a HK brokerage account, no leverage, and HK market hours.
Now you can go long or short, levered, 24/7, permissionless, onchain.
Yet, the perp is only as good as what sits underneath it. So here is what sits underneath.
➊ Zhipu (HK2513)
The first pure-play AGI lab to ever IPO, spun out of Tsinghua in 2019. Listed on Jan 8th at HK$116.20, up roughly 20-25x at the June 22 high, and currently sitting at a $120B to $150B valuation, depending on the day’s swing (note that free float is under 6%, and this stuff is volatile).
But WTF am I so excited about this?
The answer is easy:
GLM-5.2, a 753B-parameter MoE (~40B active), and with 1M context, released mid-June under a permissive MIT license.
The benchmarks:
- SWE-bench Pro: 62.1, ahead of GPT-5.5 at 58.6
- FrontierSWE: 74.4, one point behind Claude Opus 4.8 at 75.1
- Terminal-Bench: 81.0
- AIME 2026: 99.2
All at roughly one sixth the cost of @claudeai Opus. By independent confirmation it is the strongest open-weight coding model available right now, and the catalyst that the US export controls cutting Fable 5 and Mythos access for foreign nationals is hard to overstate.
Every team outside the US that just lost frontier closed-weight access was handed a near-frontier open-weight replacement at a sixth of the price.
The improved accessibility via HIP-3, the general AI hype and IPO szn are strong attention boosters here imo + if they keep shipping, I rly believe there is still meaningful upside in this.
However, 2025 revenue was only ~$100M (up ~132%) against a ~$650M net loss. At ~$120B that is north of 1,000x sales. This is priced on scarcity and capability, not cash flow, so not necessarily "cheap" per se.
➋ MiniMax (HK0100)
Shanghai lab founded in 2022 by an ex-SenseTime team, backed by Alibaba, Tencent and miHoYo. Listed Jan 9 at HK$165, doubled on debut (+109 percent, public tranche oversubscribed 1,837x), peaked at HK$1,330 in March, now sitting at around ~$20B.
It ships the open-weight M-series plus Hailuo video and audio. The current flagship, M2.7 (open-sourced March 18), is the sleeper in its basket:
- SWE-Pro: 56.22 percent, matching GPT-5.3-Codex and near Opus
- SWE-bench Verified: 78 percent
- Terminal-Bench 2: 57.0 percent,
- VIBE-Pro: 55.6 percent
- GDPval-AA: 1495 ELO, the highest of any open-weight model, ahead of GPT-5.3
And it runs roughly 50 to 60x cheaper than @AnthropicAI's Claude Opus per token. In one third-party head-to-head it delivered about 90 percent of Opus 4.6’s quality at about 7 percent of the cost.
Yet: ~$79M 2025 revenue (up ~159 percent), ARR over $300M with 300M+ users, 70% of revenue overseas(!), against a ~$250M adjusted loss. At ~$20B that is ~250x sales. Same scarcity premium as Z AI, smaller absolute number(s).
➌ The Context (China vs. USA)
Listed Chinese frontier versus US private frontier:
- Zhipu: ~$120B to $150B
- MiniMax: ~$20B
= Combined: ~$140B to $175B
vs.
- Anthropic: $965B (private, IPO coming up)
- OpenAI: $852B (private, IPO coming up)
- SpaceX: over $2T (public, including xAI and Grok)
So, $ZHIPU and $MINIMAX together are less than a fifth of Anthropic alone, while shipping open-weight models within one to a few points of the US closed frontier on coding.
On benchmarked capability per dollar of market cap, this is the cheapest frontier exposure that exists, and as of today it trades onchain with leverage.
➍ Conclusion
Three things line up:
- One, the open-weight frontier is re-rating as a category, and these are the only two pure-play listed names in it.
- Two, the export-control regime keeps handing open-weight labs demand they did not have to earn.
- Three, HIP-3 just removed every access barrier, so the marginal onchain buyer can finally express this view, and OI is already building (Zhipu at $9.4M in 7 hours).
But be aware of the risks:
These are some of the probably most volatile equities on earth. ZHIPUswings 25 to 30 percent a day on a sub-6 percent float, the multiples are massive (Zhipu north of 1,000x sales, MiniMax ~250x), and hence any cooling of the AI bid hits these first and hardest.
Also: Both have cornerstone lock-ups expiring around July 8, dumping a large slug of new float onto the market right as the hype is picking up. Be careful with leverage and size accordingly.
QUOTING @zerokn0wledge_
Very interesting. HIP-3 dominator @tradexyz grabbing the $ZHIPU ticker means a perp tracking @Zai_org's Hong Kong-listed shares is coming to Hyperliquid. Why this matters? $ZHIPU is a Tsinghua spinout and basically the first pure-play LLM lab that went public (hkex 2513). It's
Very interesting. HIP-3 dominator @tradexyz grabbing the $ZHIPU ticker means a perp tracking @Zai_org's Hong Kong-listed shares is coming to Hyperliquid. Why this matters? $ZHIPU is a Tsinghua spinout and basically the first pure-play LLM lab that went public (hkex 2513). It's
3 more images on X