China's Zhipu AI Launches GLM-5 with 30% Price Increase—As Stock Jumps 34%

Last month, the senior AI architect Lynn Cole made a prediction about Zhipu's next model in an interview with Techloy. "4.8 is going to be the best model on the market if it improves as much between 4.7 and 4.8 as it did from 4.6 to 4.7," she said. 

Zhipu AI skipped 4.8 entirely. On February 11, the company launched GLM-5, a 744-billion-parameter model that ranks first among open-source models on multiple benchmarks. On SWE-bench Verified, which measures how well models fix real GitHub issues, GLM-5 achieved a score of 77.8, outperforming Gemini 3 Pro (76.2) and approaching Claude Opus 4.6 (80.9). 

The same day, Zhipu raised prices for its GLM Coding Plan by 30%. The company noted on X, "GLM Coding Plan has seen strong growth in users and usage. To sustain service quality, we've been investing heavily in compute and model optimization. To reflect these rising costs, we're adjusting GLM Coding Plan pricing effective February 11, 2026". 

Bloomberg reported Zhipu's shares rose 34% in Hong Kong following the announcement on Thursday, reflecting market confidence in the technical achievement.  

The Performance Claims 

GLM-5 achieves performance alignment with Claude Opus 4.5 in software engineering tasks, reaching the highest scores among open-weight models across widely recognized industry benchmarks. The model scales from GLM-4.7's 355 billion parameters to 744 billion, with 40 billion active parameters in its Mixture-of-Experts architecture. 

GLM-5 achieved a -1 score on the AA-Omniscience Index, representing a 35-point improvement in hallucination rates. It now leads the industry in knowing when to abstain from answering rather than fabricating information—a critical capability for production environments. 

Early users report cleaner code generation, better handling of multi-file projects, and improved backend system design. The model's "Agent Mode" can convert prompts directly into production-ready documents including .docx, .pdf, and .xlsx files without manual formatting. 

However, Lukas Petersson, co-founder of Andon Labs, the lab that independently conducted the Vending Bench 2 (a benchmark that measures long-term operational capability) noted a gap. GLM-5 shows less situational awareness in complex tasks compared to Claude—a reminder that synthetic benchmarks don't always capture production reliability. 

“After hours of reading GLM-5 traces: an incredibly effective model, but far less situationally aware. Achieves goals via aggressive tactics but doesn't reason about its situation or leverage experience. This is scary. This is how you get a paperclip maximizer,” Lukas shared on X. 

The Price Adjustment 

Zhipu's X post detailed the changes: "First-purchase discounts removed; quarterly and annual discounts added. Prices increase for Lite and Max plan. Existing subscribers keep their current pricing." 

The promotional $3 entry point that made GLM-4.7 accessible is gone. New users face adjusted pricing across subscription tiers. For GLM-5 API pricing, the commonly listed baseline is $1.00 per 1M input tokens and $3.20 per 1M output tokens.  

That pricing remains significantly cheaper than Claude Opus 4.6, which costs $5 per million input tokens and $25 per million output tokens. You can also try the model for free on z.ai.  

 The Infrastructure Question

Deploying GLM-5 requires 1,490GB of memory—roughly double GLM-4.7's footprint. Running it locally means datacenter infrastructure or expensive multi-GPU setups. For Zhipu running it as a cloud service, operational costs doubled while user demand surged. 

The company's warning about tight compute signals strain before adding a model twice the size of its predecessor. For developers relying on Zhipu's API, that translates to potential rate limiting or degraded response times as the company scales infrastructure. 

"Compute is very tight. Even before the GLM-5 launch, we were pushing every chip to its limit just to serve inference," the company posted on X. 

What Changed from GLM-4.7 

Two months ago, Zhipu seemed to challenge OpenAI and Anthropic premium AI pricing tier with a simple proposition: frontier performance at $3 instead of $200. 

GLM-5 validates the technical achievement is sustainable. Cole's prediction about market-disrupting performance was accurate, even if the version number differed. The model delivers on benchmarks that matter for coding workflows. 

But the price increase follows the same cost-pressure trajectory  that Zhipu AI set out to challenge. Frontier AI at scale requires massive compute, regardless of who provides it. 

For developers who adopted GLM based on cost, the value proposition is shifting. GLM-5 delivers meaningful performance improvements—better code quality, stronger agentic capabilities, reduced hallucinations.  

Teams with hardware to run it locally get frontier-level performance with complete control. The model weights are available under MIT license on Hugging Face and ModelScope. 

Teams using Zhipu's cloud API face a different calculation. Strong capabilities remain cheaper than Claude or ChatGPT Pro. But infrastructure strain creates uncertainty that didn't exist with GLM-4.7's more modest resource requirements.