Knowledge Atlas GLM 5.2 ranks 3rd globally as Jefferies flags 94x P/ARR

Knowledge Atlas' GLM 5.2 became the first Chinese AI model to rank among the world's top three, scoring between Anthropic's Claude Opus 4.7 and 4.8 on long-horizon task benchmarks, according to the Artificial Analysis Intelligence Index.

"GLM 5.2 ranking third globally is a first for Chinese AI models," Jefferies analysts said in a research report. The model ranked fourth in coding capabilities and second in agent capabilities.

The model supports a 1-million-token context window, purpose-built for long-horizon tasks, and runs on domestic computing platforms from launch. It is released under the MIT open-source license with no regional restrictions. Multiple long-horizon task benchmarks show GLM 5.2's performance sits between Claude Opus 4.7 and 4.8, making it the highest-ranked open-source model globally.

Despite the technical achievement, Jefferies cautioned that maintaining model leadership will be challenging. The broker expects Knowledge Atlas to face continued shortages in high-end inference computing power, potentially limiting its ability to serve enterprise demand for long-context agent workloads. The stock surged 12.53 percent on the news, with short selling reaching $87.83 million, or 1.475 percent of turnover.

Valuation Gap Widens as Revenue Multiple Hits 94x

Jefferies flagged a stark valuation disconnect. Based on Knowledge Atlas' guidance of achieving $1 billion in annualized recurring revenue by end-2026, the stock trades at 94 times that forward ARR. By comparison, Anthropic — whose Claude models consistently rank at or near the top of global benchmarks — trades at roughly 18 times ARR.

"US restrictions on access to Anthropic Fable 5 may be difficult to sustain," Jefferies noted, arguing that developers have multiple alternatives and open-source models to choose from, making it unlikely the ban will generate substantial revenue growth for Knowledge Atlas.

Domestic Compute Strategy Faces Enterprise Test

Knowledge Atlas designed GLM 5.2 to run on domestic computing power platforms from day one, a strategic choice that insulates it from US chip export restrictions. But Jefferies warned that high-end inference compute shortages could constrain the company's ability to meet enterprise demand for the model's signature feature — stable handling of million-token context windows for long-horizon agent workloads.

The company's open-source strategy under the MIT license, with no geographic restrictions, aims to drive adoption across global developer communities. However, the tension between open-source distribution and monetization remains unresolved, with the 94x ARR multiple suggesting investors are pricing in rapid revenue conversion that may not materialize if compute bottlenecks persist.

Knowledge Atlas shares, which have more than doubled year-to-date, now price in a level of revenue growth that exceeds even the most richly valued US AI peers. If the company delivers on its $1 billion ARR target by end-2026, the current multiple could prove justified. But any miss — whether from compute constraints, competitive pressure from other open-source models, or slower enterprise adoption — would leave the stock exposed to a sharp re-rating. CMSI, in a separate note, maintained that the China AI internet sector remains attractive, naming Knowledge Atlas among its top picks alongside Alibaba, Kuaishou, and JOYY.

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