Chipu‘s GLM 5.2 artificial intelligence model debuted last week with a Silicon Valley-like buzz following the launch of DeepSeek last year.
The Chinese AI startup’s new open-source model has outperformed all other open releases and now sits within one percentage point of Anthropic’s Opus 4.8 in one high-profile agent benchmark, at about one-fifth the cost.
As developers continue to flock to us, OpenRouter token traffic is growing faster than it has since the release of DeepSeek V4 in April. And unlike DeepSeek, which the market ultimately dismissed as a one-time chatbot shock, GLM 5.2 excels at the kind of agent work that companies are racing to automate: planning, coding, testing, and looping.
AI spending on tokens, which are measures of data processed and generated by AI models, has become unexpectedly high, and businesses that have been hit are now looking for ways to make the most of their funds.
The most important metric is going to be intelligence per dollar, and Zhipu’s cheap but sufficient model is an attractive answer.
“I’m always amazed at how quickly open source catches up,” Harvey co-founder Gabe Pereyra told CNBC. “With GLM 5.2, you’re seeing the first model that’s really competitive with some of these closed-source frontier models.”
But the real story beneath the price tag is the open source AI moment.
GLM 5.2 is free to download, tweak, and run on a company’s own servers, making access both unstable and putting pricing pressure on Frontier Labs. Anthropic had to retire its Fable Mythos class of models following orders from the Trump administration, and OpenAI announced Friday that it would restrict its GPT 5.6 models at the government’s request.
With federal oversight, a model that no one can undo is looking increasingly like a safe option.
