Zhipu AI's Trillion-Dollar Surge: The Perfect Storm of Open-Source Excellence and US Export Controls

• SWE-bench Pro: 62.1, beating GPT-5.5's 58.6 and trailing Claude Opus 4.8 by roughly 7 points
• FrontierSWE (Dominance): 74.4, surpassing GPT-5.5's 72.6 — within 0.7 points of Claude Opus 4.8's 75.1
• Terminal-Bench 2.1: 81.0, landing within a few points of Claude Opus 4.8's 85.0
• GPQA-Diamond: 91.2 — a reasoning benchmark where open-weight models rarely exceed 85
• Code Arena Frontend: ranked second globally, first among all open-weight models

Independent benchmarker Artificial Analysis ranked GLM-5.2 as the best open-weight model in the world and fourth overall behind only closed frontier models .

The 82% Price Gap

Zhipu priced GLM-5.2 at $1.40 per million input tokens and $4.40 per million output tokens — roughly one-sixth the blended cost of GPT-5.5 . By comparison, Anthropic's Claude Opus 4.6 cost $5 per million input tokens and $25 per million output tokens as of early 2026 .

The cost advantage translated directly into real-world economics. Independent security firm Semgrep found that GLM-5.2 detected web vulnerabilities at roughly $0.17 per finding, while Claude Code cost upwards of $0.53 per finding — a 3x efficiency gap .

The Open-Source Effect

Released under the permissive MIT license with open weights, GLM-5.2 was rapidly self-hosted and offered by multiple cloud providers within days . Developers who wanted to avoid sending proprietary code through a US-based API could download the model weights and run them on their own infrastructure — a freedom no closed-source provider could match .

Zhipu's GLM Coding Plan started at roughly $10/month for the Lite tier, compared to $20/month for comparable access from US rivals . By June 21, independent benchmarks had confirmed GLM-5.2 as the top open-weight coding model, and the adoption wave was well underway .

The US Export Ban on Anthropic: The Catalyst

On June 12, 2026, the US Commerce Department issued an emergency export-control directive barring foreign nationals from using Anthropic's newest models — Claude Mythos 5 and Claude Fable 5 — citing national security concerns . The directive was received by Anthropic at 5:21 p.m. ET that day .

Because Anthropic could not reliably separate US citizens from non-citizens in its user base, the company chose to disable both models globally, effective immediately . The models, which had debuted only three days earlier on June 9, disappeared from international markets overnight .

Commerce Secretary Howard Lutnick later said the action was taken because officials feared the models could be "deployed by military intelligence users in China, Russia or other countries of concern" . A Semafor report also indicated the ban was partly triggered by suspicions that a China-linked group had accessed Mythos .

Anthropic sent top engineers to the White House to negotiate a reversal, but as of late June, the models remained offline for all non-US users .

The Perfect Timing: GLM-5.2 Launches as Claude Goes Dark

GLM-5.2 was released publicly around June 13–17, 2026 — the very same week Anthropic's top models were taken offline . Global developers who relied on Claude's top-tier coding capabilities were suddenly locked out, and an open-source, MIT-licensed alternative that benchmarked within 1–3% of Claude Opus 4.8 on key coding metrics was sitting there, free to download and self-host .

This was not lost on investors. On June 22, the first trading day after the Dragon Boat Festival holiday, Zhipu's shares went parabolic, surging as much as 40% to an intraday high of HK$2,980 . At the close, the company's market capitalization reached HK$1.1 trillion — approximately half of Alibaba's, twice Meituan's, and three times JD.com's .

The "DeepSeek 2.0 Moment"

Analysts and media explicitly framed the confluence as a "DeepSeek 2.0 moment" — a repeat of the January 2025 DeepSeek shock, where a Chinese open-source model suddenly erased the US closed-source premium and triggered a massive re-rating of Chinese AI stocks .

JPMorgan projected an over 534% revenue surge for Zhipu in 2026 and revised its profit forecast to 2028, reversing a previous net-loss prediction . The bank had upgraded Zhipu's target price to HK$1,800 — a level the stock had already exceeded before the Anthropic ban fully took effect .

The Ironic Validation

US cybersecurity executives and industry experts warned that the restrictions on Anthropic could "help US adversaries more than it hurts them" by driving global demand toward Chinese open-source alternatives . That is precisely what happened: Zhipu went from a ~HK$51 billion IPO in January to a ~HK$1 trillion market cap in six months, fueled by a technical product win and a geopolitical supply shock .

As one analysis put it: "With Zhipu becoming the first large model enterprise with a trillion-dollar market value, this curve is constantly being validated" .