Featherless.ai & Z.ai launch GLM 5.2 access worldwide

Featherless.ai has partnered with Z.ai to offer global access to the GLM 5.2 artificial intelligence model, a 744-billion-parameter open-source system for software development.

Under the arrangement, GLM 5.2 is available through Featherless.ai's OpenAI-compatible application programming interface. Featherless.ai hosts the model directly, rather than requiring customers to provision their own graphics processing infrastructure. The model is available in FP8, with a context window of up to 256,000 tokens on public cloud deployments and up to 1 million tokens on private cloud deployments.

The launch comes as access to advanced AI systems has become more politically sensitive. Featherless.ai linked the timing to new US export controls on Anthropic's Fable 5, arguing that open-source alternatives are likely to draw more attention when companies fear access to closed models could be restricted.

GLM 5.2 is positioned as a coding-focused model for organisations seeking an alternative to proprietary systems for software engineering work. It is released under the MIT licence, which allows broad commercial use.

According to the companies, Z.ai's system uses a Mixture-of-Experts design with 744 billion parameters, of which 39 billion are active per token. Featherless.ai said the model has a similar physical size to its predecessor, GLM 5.1, while improving results through training focused on coding agents and a longer context window.

Among the technical changes cited are an IndexShare feature in every fourth sparse attention layer, intended to maintain quality during extended coding sessions, and a revised multi-token prediction layer that improves speculative decoding acceptance by about 20%. The model also includes High and Max settings that let users adjust the trade-off between reasoning depth and response speed.

Benchmark claims

Featherless.ai said GLM 5.2 performs strongly on a series of software and reasoning benchmarks. It described the model as the top-ranked open-source system on long-horizon programming tests including FrontierSWE, PostTrainBench and SWE-Marathon, and said it competes with closed models such as Claude Opus and GPT-5.5.

The company published a range of score improvements from earlier model generations. Terminal-Bench 2.1 rose to 81.0 from 63.5, while SWE-bench Pro increased to 62.1 from 58.4. On more complex software development tests, FrontierSWE increased to 74.4 from 30.5 and SWE-Marathon rose to 13.0 from 1.0.

It also cited gains on reasoning tasks. AIME 2026 rose to 99.2 from 95.3, while GPQA-Diamond increased to 91.2 from 86.2.

AMD focus

Featherless.ai is also introducing a private cloud deployment that pairs GLM 5.2 with AMD infrastructure. It said it is the only platform to have optimised GLM 5.2 to run natively on AMD hardware, framing that as a way for customers to avoid supply constraints tied to Nvidia chips.

That pitch comes as access to advanced chips remains a strategic concern for AI infrastructure providers and enterprise buyers. By emphasising AMD-based deployments, Featherless.ai is trying to distinguish itself in the crowded market for hosted open models, where many providers still rely heavily on Nvidia hardware.

Its hosted approach also includes no logs and data hosting in both Europe and the US, aimed at addressing customer concerns around infrastructure management and geographic deployment options. Featherless.ai did not disclose pricing for access to GLM 5.2 or for private cloud deployments.

Featherless.ai describes itself as a serverless inference platform with a catalogue of more than 30,000 open models. It recently raised USD $20 million in a Series A funding round led by AMD Ventures and Airbus Ventures, with participation from BMW i Ventures.

Commenting on the wider market backdrop, Eugene Cheah, Chief Executive Officer and Co-Founder of Featherless.ai, said: "Any attempt to enforce tight export controls or closed architectures as a way to keep the frontier features in check will only result in the development of an alternative. In the situation where an engineering team or company feels that the access might be revoked suddenly, there is an incentive to switch to technologies that they can operate and change entirely independently of anyone else. That's precisely why open source technology evolves so quickly."