GLM 5.2 pairs a 1M-token context with long-horizon agents, and Fireworks reproduces its GPQA-Diamond score on its own stack
Synopsis
Z.ai released GLM 5.2, a flagship built for long-horizon agentic tasks on a solid 1M-token context and positioned by its published benchmarks as the strongest open-source coding model. Rather than forwarding vendor numbers, Fireworks ran the open weights on its own GPUs and inference engine and scored 91.4% (181/198) on GPQA-Diamond in high reasoning mode, matching Z.ai's reported 91.2%. The model ships under an MIT license and is live on Fireworks for developers and LLM researchers.
Interpretation
GLM 5.2 targets long-horizon agentic work, and its headline feature is a solid 1M-token context window meant to let an agent run unattended for extended periods. The prior condition is that a 1M context is easy to claim but much harder to keep reliable under real engineering pressure, as Z.ai's technical blog states. Fireworks notes that pushing context to 1M moves the bottleneck from compute to KV-cache capacity, kernel overhead and CPU-side scheduling, so the window matters only if it stays coherent through hours of autonomous running.
Fireworks ran the GLM 5.2 open weights on its own GPUs and inference engine, independently checking the reasoning score Z.ai published. Vendor benchmarks usually run on the provider's own infrastructure and harness, which is why cross-model comparisons stay contentious, so Fireworks did not take the vendor chart on faith. On GPQA-Diamond, GLM 5.2 scored 91.4% (181/198) on Fireworks' engine and GPUs in high reasoning mode, against Z.ai's reported 91.2%.
GLM 5.2 ships under an MIT license, permitting commercial use, modification and redistribution with no copyleft strings and no regional limits. It arrives within a fast run of open releases in the past week alongside new flagships from Kimi, Qwen, DeepSeek and MiniMax. Fireworks lists it beside those same-week open flagships and states the model is live on its platform for developers and LLM researchers.
Perspective
The demonstrated scope is GLM 5.2 running GPQA-Diamond in high reasoning mode on Fireworks' own GPUs and inference engine at 91.4% (181/198), and the model being live on Fireworks through the Playground, serverless per-token access, and OpenAI- and Anthropic-compatible endpoints.
Fireworks states that public benchmarks answer a general question rather than a specific user's, and that a leaderboard winner can still lose on a given codebase, prompts and latency budget, so the reliable choice is to test candidates on the work actually shipped. The coding positioning of GLM 5.2 rests on Z.ai's published benchmark results.