Zhipu's H1 Revenue Surges 400% to 954M RMB as GLM-6 Bets on Recursive Self-Improvement

Zhipu AI released its H1 2026 results and next-gen model roadmap: H1 revenue hit 954M RMB (400% YoY growth), exceeding all of 2025, while MaaS annualized revenue jumped from $250M in March to $2B in August, and the API business is now profitable. GLM-6.0 will be trained by GLM-5.3 via recursive self-improvement (RSI), one of the most aggressive automated R&D bets among Chinese model labs.

Confirmed

H1 2026 revenue of 954M RMB, up 400% YoY, exceeding full-year 2025
API/open platform revenue up 27x YoY, accounting for 86.5% of revenue; MaaS annualized revenue rose from $250M (March) to $2B (August); August monthly revenue exceeded the entire H1 API total; API business is profitable
Token usage up over 40x this year; inference costs down 80% while average API pricing increased
GLM-6.0 will be trained by GLM-5.3 and fully adopt the RSI approach, with models self-training in environments built by prior generations
Listing documents (pp. 19-20) describe an early RSI form: the GLM-5.3 Agent helped engineers optimize GLM-5.3-Flash inference, delivering a 300% end-to-end performance gain on domestic chips
The earnings call also predicted a shift from single agents to multi-agent collaboration, and from customers buying tokens to buying tasks

Why it matters

Zhipu's pivot from private deployment to MaaS is complete and profitable
Per Emad Mostaque, GLM 5.3's pretraining dates to early this year; amid internet data exhaustion the company is massively scaling data environments, making GLM 6.0's full RSI bet an aggressive automated R&D route; Tang Jie views RSI as a core strategic direction

2026-09-01 ~ 2026-09-02 · 6 related posts

Primary sources

Zhipu filing reveals GLM agent boosted inference speed by 300% — imjustnewatai · 2026-09-01 09:05 (UTC+08:00)
Zhipu AI H1 Report: Revenue up 400%, API unit costs down 80% — 赛博禅心 · 2026-09-01 01:12 (UTC+08:00)

[source] Zhipu AI H1 Report: Revenue up 400%, API unit costs down 80% — 赛博禅心 · 2026-09-01
Zhipu API ARR hits $2B; inference costs drop 80% on domestic chips — xeophon · 2026-09-01
Zhipu earnings: $2B ARR, GLM 6.0 to fully embrace RSI, says Mostaque — ccerrato147 · 2026-09-01
[source] Zhipu filing reveals GLM agent boosted inference speed by 300% — imjustnewatai · 2026-09-01
Zhipu Announces Recursive Self-Improvement for GLM-6, Hits $2B ARR — koltregaskes · 2026-09-02
Zhipu Earnings: GLM Self-Iteration, Shift from Tokens to Task Outcomes — rohanpaul_ai · 2026-09-02