Zhipu Surges 33% as Z.ai Ships GLM-5.2

What drove Zhipu's stock surge on June 15?

Zhipu shares jumped as much as 48% on June 15, 2026, before settling 33% higher at roughly HK$1,461 ($186). JPMorgan maintained its overweight rating and raised its price target to HK$1,400 from HK$950, citing the firm's model visibility and pricing power. The bank simultaneously downgraded domestic rival MiniMax, according to a Bloomberg report cited by CNBC.

MiniMax shares also rose — up 7.4% on the same day.

What did Bank of America do?

Bank of America separately initiated coverage on both companies with "buy" ratings. It set a target of HK$1,250 for Zhipu and HK$500 for MiniMax. Both initiations landed on the same Monday as the JPMorgan move, adding further Wall Street attention to Chinese AI stocks.

How does Anthropic factor in?

The share moves came as Washington tightened curbs on foreign access to powerful U.S. AI models. CNBC reported that Anthropic's restrictions on international access are part of the backdrop pushing investors toward Chinese AI alternatives. For teams evaluating sovereign AI strategies, this dynamic is increasingly relevant.

What is GLM-5.2?

GLM-5.2 is the latest large language model from Z.ai — the lab arm of Zhipu — and the third major release in the GLM-5 line. It launched on June 13, 2026. The model's headline spec is a 1,000,000-token context window, labeled glm-5.2[1m] in Z.ai's own configuration. Each response can return up to 131,072 output tokens.

You might also like

Sen. Warren Demands SEC Halt SpaceX IPO

AI Productivity Boom Won't Fix the US Deficit

SpaceX Stock Surges Past Amazon After Record IPO

US Holds Back DeepSeek Trade Blacklist

That context window is roughly 5x larger than GLM-5.1's ~200,000-token window, according to MarkTechPost.

How does GLM-5.2 compare to GLM-5.1?

What are the two thinking-effort levels?

GLM-5.2 adds two thinking-effort settings: High and Max. Z.ai recommends Max effort for complex, multi-step coding work. In Claude Code, the /effort command controls this. The xhigh, max, and ultracode options all map to GLM-5.2's Max effort level.

How does GLM-5.2 pricing compare to Claude Opus 4.8?

Cost is the clearest verifiable difference. The GLM line has historically priced around $1 per million input tokens and $3.20 per million output tokens. Claude Opus 4.8 runs $5 per million input and $25 per million output — a 5x to 8x gap on output tokens alone, according to Lushbinary's comparison.

Z.ai did not publish standalone GLM-5.2 API pricing at launch. The company said API access would follow within about a week. The GLM Coding Plan subscription is the day-one access path.

Here's what we know so far on the benchmark front: Z.ai published zero benchmark scores for GLM-5.2 at launch. There is no SWE-bench, Terminal-Bench, or Code Arena number. GLM-5.1 scored 58.4 on SWE-bench Pro; Claude Opus 4.8 posted 88.6% on SWE-Bench Verified. GLM-5.2's performance relative to either remains unmeasured.

What is the GLM-5.2 license?

GLM-5.2 ships under an MIT license. Z.ai said open weights would be released approximately one week after the June 13 launch. MIT licensing allows self-hosting, fine-tuning, quantization, and air-gapped deployment. Claude Opus 4.8 and GPT-5.5 are closed and API-only — a hard blocker for teams with strict data-residency requirements.

For developers tracking AI coding infrastructure, the self-hosting option matters when frontier API access is disrupted.

What is the GLM-5 model lineage?

Z.ai has shipped four flagship-tier releases in roughly four months:

• GLM-5 — February 11, 2026
• GLM-5-Turbo — March 15, 2026
• GLM-5.1 — April 7, 2026
• GLM-5.2 — June 13, 2026

The GLM-5 base is reported by the community to be a 744-billion-parameter Mixture-of-Experts model that activates 40 billion parameters per token. Z.ai did not specify GLM-5.2's architecture in its launch materials.

What use cases does Z.ai target with GLM-5.2?

Z.ai points to three primary use cases:

• Whole-repository refactors — the 1M window holds an entire mid-sized codebase, including source files, tests, and configuration, without constant re-fetching.
• Long-horizon agent runs — GLM-5.1 sustained roughly 1,700 agent steps in one session and ran autonomous loops for up to eight hours. GLM-5.2 targets the same trajectory.
• Drop-in Claude Code replacement — swap the base URL and model identifier only, keeping an existing agent harness intact.

For teams building agentic coding workflows, the 1M window means the agent can remember decisions made hundreds of steps earlier in a long task.

The open-weight release of GLM-5.2 weights is the next confirmed milestone Z.ai has announced, expected approximately one week after the June 13 launch date.