Zhipu AI has quietly raised the default concurrency ceiling on its GLM API, letting developers run more requests in parallel without filing a support ticket first. It's a fix for a problem the company largely created for itself.
If you have tried to build anything serious on top of GLM this year, you already know the wall you hit. The models are competitive, the pricing undercuts most Western rivals, and the open weights on Hugging Face have pulled in a real developer base. But the moment your application tries to fire off more than a couple of requests at once, the API pushes back. Z.ai, the Beijing-founded, Hong Kong-listed company behind GLM, has now raised the default concurrency capacity for GLM API users, according to the company's own developer documentation. The change applies to API access specifically. It doesn't touch the separate GLM Coding Plan.
That distinction matters more than it sounds. Z.ai frames the upgrade as a response to growing developer demand and more real-world applications, and says the goal is reliable, production-ready infrastructure with a smoother development experience. Developers who still need more throughput than the new default allows can request additional capacity through a form on Z.ai's site rather than waiting on a manual review. For teams running agentic workflows, where a single task can spin off a dozen parallel tool calls, that's not a cosmetic change. It's the difference between an API you can build a product on and one you have to work around.
Z.ai didn't frame this as damage control, but the timing lines up with a rough stretch for GLM's reputation among developers. An open issue on the zai-org GLM-5 GitHub repository, filed as a critical bug, describes the GLM-5.2 API as effectively unusable for two consecutive days because of severe rate limiting. Separately, a bug report filed against the opencode project details GLM Coding Plan Pro, the $15-a-month tier, refusing more than a single in-flight request at a time, a limit the reporter says is nowhere in Z.ai's published pricing. Developers on Reddit have described similar throttling during normal working hours, and some say multi-agent tools that spawn background tasks are effectively unusable on the platform.
None of that is disclosed in Z.ai's announcement. But it explains why letting developers run more requests in parallel reads less like a feature drop and more like an overdue repair. The company is explicit that Coding Plan users are excluded from this particular fix, which means the exact workflow drawing the loudest complaints, GLM Coding Plan Pro subscribers running agentic coding tools, gets nothing from this update. If you're paying for GLM through the API directly, your ceiling just moved. If you're paying through the Coding Plan, you're waiting on a separate fix that Z.ai hasn't announced.
Z.ai has real incentive to get this right. The company listed on the Hong Kong Stock Exchange on January 8, 2026, raising $558 million at a $7.1 billion valuation, and its shares have since climbed nearly 1,000%, according to Forbes, pushing chairman Liu Debing's fortune to roughly $22.4 billion. GLM-5.2, released in June with a 1 million token context window and 744 billion total parameters, has been positioned by Z.ai as proof the company can compete with DeepSeek and Alibaba's Qwen on frontier open-weight models while also running commercial infrastructure at scale. The GLM Coding Plan alone has drawn more than 150,000 users. That's a lot of paying customers to leave stuck behind a concurrency wall while the API tier gets fixed first.
There's a strategic logic to sequencing it this way. API access is where enterprise customers and larger integrations sit, and it's the tier most exposed to churn if a competitor like Moonshot's Kimi or DeepSeek offers a smoother pipe. Frankly, the Coding Plan's individual developers have fewer alternatives at the same price point, so they're more likely to tolerate the wait. That's a reasonable business bet. It's also the kind of bet that erodes goodwill if it drags on much longer, especially with an open GitHub issue tracking the outage in public.
Open weights got Z.ai noticed. Reliable infrastructure is what keeps developers building on the hosted API instead of just downloading the model and running it themselves. This concurrency upgrade is a small, technical step in that direction. Whether Z.ai extends the same fix to the Coding Plan, and how fast, will say a lot more about the company's priorities than the announcement itself does.
Also read: What Is an AI Moat and Why Most LLM Wrapper Startups Have None • Samsung's Record Profit Couldn't Stop Its Stock From Cratering This Week • OpenAI clears government hurdle to launch GPT-5.6 for everyone Thursday
