Z.ai open-sources GLM-5.2, beating GPT-5.5 on coding at 1/6th cost

Z.ai's GLM-5.2 matches or beats OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8 on key coding benchmarks at a fraction of the cost, while its MIT open-source license lets enterprises bypass geographic restrictions.

Z.ai released GLM-5.2, a 753-billion-parameter open-weights model that beats GPT-5.5 on multiple long-horizon coding benchmarks while charging 1/6th the API price, intensifying competition in the frontier AI market.

"GLM-5.2 is the first open-weights model to cross 80% on Terminal-Bench, and beats every other open model available," the team behind Cline IDE said on X, announcing day-one integration with the open-source coding environment.

The model scores 62.1 on SWE-bench Pro, surpassing GPT-5.5's 58.6 and its predecessor GLM-5.1's 58.4. On FrontierSWE, which tests long-horizon task completion, it hits 74.4%, trailing Claude Opus 4.8's 75.1% by less than one point. Its 1-million-token context window — expanded from 200,000 tokens in GLM-5.1 — is powered by a new IndexShare architecture that reuses the same indexer across every four sparse attention layers, reducing per-token compute FLOPs by 2.9 times at maximum context length.

API pricing starts at $1.40 per million input tokens and $4.40 per million output tokens, compared with GPT-5.5 at $5.00 and $30.00. The MIT open-source license allows enterprises to host the model locally, eliminating vendor lock-in and bypassing the geographic restrictions that recently blocked foreign access to Anthropic's Claude Fable 5 after a US export control directive.

On the Design Arena crowdsourced benchmark, GLM-5.2 took first place with an ELO score of 1,360, beating even Claude Fable 5. On Code Arena, a front-end development evaluation system involving blind tests by millions of global users, it achieved the top ranking among all available models. The model also introduces selectable "Thinking Modes" — a "Max" setting pushes peak intelligence using roughly 85,000 output tokens per task, while "High" halves token consumption with minimal performance loss.

The developer ecosystem has responded quickly. Kilo Code, Cline, and Eigent AI all confirmed day-one support. Subscription tiers for the GLM Coding Plan start at $12.60 per month for the Lite tier, $50.40 for Pro, and $112.00 for Max, each targeting progressively heavier repository workloads.

Knowledge Atlas (02513.HK), Z.ai's Hong Kong-listed parent, rose 0.7% on the announcement, with short-selling volume reaching $92.75 million — a 2% short ratio that suggests significant trader positioning. CICC raised its price target 39% to HKD 1,250, maintaining an Outperform rating, while JPMorgan increased its target 47% to HKD 1,400, citing the company's leading position in coding capabilities.

The competitive implications extend beyond benchmarks. With DeepSeek-V4-Pro charging $0.87 per million output tokens and GLM-5.2 at $4.40, the gap between open-weights and proprietary models has narrowed to the point where leading labs may be operating at "90%+ margins," as AI observer Lisan al Gaib argued on X. For enterprises, the ability to run frontier-level AI on sovereign infrastructure under an unrestricted MIT license removes a key barrier to adoption — particularly for organizations in jurisdictions affected by US export controls.

This article is for informational purposes only and does not constitute investment advice.