Z.ai logo

Z.ai

NativeAI NativePublic · 2513.HK

China's most research-credible foundation model company — and one of the few outside the US-Europe axis worth watching.

Website
z.ai
Founded
2019
CEO
Zhang Peng
HQ
Beijing, China
Stage
Public
Revenue
$100M–$500M
Employees
201-500
Flag a data issue
62.9Bluescore
Momentum61.0
Market MapDeep Tech
Rank change (7d)▼ 23

Executive Summary

Updated Nov 12, 2026

What they do

  • Builds the GLM family of frontier large language and multimodal foundation models
  • Releases open-weight GLM models under MIT licence alongside a paid API
  • Sells the GLM Coding Plan subscription for agentic coding inside Claude Code and Cursor

Why it matters

  • GLM-5.3 pairs top coding and agentic scores with low $1.40/$4.40 token pricing
  • First pure-play LLM company to go public, so its financials are disclosed openly
  • H1 2026 revenue up 399.7% to RMB 953.9M on API demand

Best fit

  • Teams wanting strong open-weight models to self-host on their own infrastructure
  • Cost-sensitive developers seeking cheaper coding-agent capacity than Western frontier providers
  • Enterprises in China or Asia needing models that run on Huawei Ascend chips

Watch-outs

  • On the US Entity List since January 2025, a compliance and procurement red flag
  • Losses of about RMB 2 billion in H1 2026, with revenue roughly 30% below estimates
  • Coding Plan users report restrictive, poorly documented rate limits

Latest News

Analyst VerdictEvaluate

Updated Nov 12, 2026

Z.ai is one of the most technically credible model makers outside the US, and its MIT-licensed GLM-5.x models are a real low-cost option for coding and agent work. Self-hosting the open weights can reduce data-residency concerns. Buyers should run a structured evaluation, and US and allied buyers in particular should check Entity List and data-sovereignty exposure before using the hosted API. Companies in regulated or government-adjacent sectors should treat legal review as a gating step.

Signal Provenance

Latest observation date and source for each signal category

Signal CategoryLast UpdatedSource
Loading…

Why this Bluescore?

6 dimensions, each scored 0–10 against all 250 companies. Each adds its weight × score to the Bluescore — see methodology

Z.ai's Bluescore of 62.9 rests most on Data & Model Moat (10.0, top 1%) and Talent & Research (8.2, top 10%). Its weakest area is Investment Momentum (0.6, 1st percentile).

In plain English

Z.ai's Bluescore of 62.9 out of 100, which ranks #29 of 247, is lifted mainly by a Data & Model Moat score of 10.0 of 10 that adds 19.0 of 19 points and a Talent & Research score of 8.2 of 10 at the 90th percentile. Investment Momentum holds it back most sharply, with a score of 0.6 of 10 at the 1st percentile, though its 10% weight limits the damage. The biggest opportunity is Business Model & Revenue, which scores 4.9 of 10 and still has 10.7 of 21 points left to earn, so improving it would move the Bluescore more than any other dimension.

Written by AI from the scores below and checked against them; updated nightly.

D1 · Business Model & Revenue (21%)
4.9
+10.3 of 21 pts · 71st pct
D2 · Product Depth (21%)
6.2
+13.1 of 21 pts · Top 22%
D3 · Data & Model Moat (19%)
10.0
+19.0 of 19 pts · Top 1%
D4 · Talent & Research (16%)
8.2
+13.1 of 16 pts · Top 10%
D5 · Customer Adoption (13%)
5.2
+6.8 of 13 pts · 50th pct
D6 · Investment Momentum (10%)
0.6
+0.6 of 10 pts · 1st pct

Traction Signals

Last Round

Public

Risks

Updated Nov 12, 2026

US Entity List designation creates export-control, procurement and reputational exposure for Western buyers
High
Hosted API routes data through a China-based provider, raising data sovereignty concerns
High
Heavy model and compute spending keeps losses large despite fast revenue growth
Medium
Limited access to US chips could slow future model scaling and inference capacity
Medium
Fierce price war with DeepSeek, Moonshot, MiniMax and Alibaba Qwen squeezes margins
Medium
Feedback