GLM 5

GLM 5 is designed for complex systems engineering and long-horizon agentic work, especially tasks that require sustained reasoning and coding rather than short isolated responses. Its architecture expands on the preceding generation to 744 billion total parameters with 40 billion active, while the training corpus grows to 28.5 trillion tokens. DeepSeek Sparse Attention is used to reduce deployment cost while retaining long-context capacity, making the model better suited to large projects and multi-step workflows.

Training combines large-scale pretraining with reinforcement-learning refinement supported by the asynchronous slime infrastructure, which improves training throughput and enables more iterative post-training work. The resulting model is described as improving substantially over the previous generation across academic benchmarks, with particularly strong results in reasoning, coding, and agentic tasks. In practice, it is a strong fit for software engineering, tool-assisted problem solving, and other workflows where maintaining context over a long sequence of actions matters.

Basetenzai-org/GLM-5glm

Open provider page Provider docs

Quick Info

Provider: Baseten
Model key: zai-org/GLM-5
Release date: Feb 12, 2026
Last updated: Feb 12, 2026
Knowledge cutoff: 2026-01
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.95
Output token cost: $3.15

Limits

Output tokens: 202,800 tokens
Context window: 202,800 tokens

Transparent token rates

Compare GLM 5 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GLM 5

Baseten

CoverageAnalysis

Artificial Analysis reports that Z.ai's GLM-5.2 is the new leading open-weights model on its Intelligence Index v4.1, scoring 51 and sitting on the Pareto frontier of intelligence vs. cost per task. The model keeps GLM-5.1's 744B-total / 40B-active parameter footprint but jumps 11 points on the Intelligence Index, ahea Key benchmark gains over GLM-5.1 include +16 points on CritPt (to 21%), +12 on HLE (to 40%), +9 on AA-LCR (to 71%), +15 on tau3 banking (to 27%), +7 on SciCode (to 50%), +16 on TerminalBench v2.1 (to 78%), and +3 on GPQA Diamond (to 89%); GLM-5.2 also leads open weights on GDPval-AA v2 with a score of 1524. The model u

Baseten

Official sourceRelease Notes

GLM 5 takes the intelligence from its predecessors and layers in long-horizon agentic capabilities and complex systems engineering. You can deploy GLM 5 in one click with our Model APIs; dedicated...

Baseten

CoverageBenchmark

DeepInfra's provider blog profiles GLM-5.2 as a 753B-total / 40B-active parameter Mixture-of-Experts reasoning model from Z.ai, released June 16, 2026, with a 1M-token context window, MIT licensing, JSON output, function calling, and English/Chinese support. It frames GLM-5.2 as notable for combining strong benchmark p The post cites Artificial Analysis's Intelligence Index score of 51 for GLM-5.2 and OpenRouter rankings placing it above ~88% of models on intelligence, ~87% on coding, and ~90% on agentic performance. DeepInfra's own benchmark tables show GLM-5.2 holding up against Qwen3.7-Max, MiniMax-M3, DeepSeek-V4-Pro, Claude Opus