DigitalOcean
CoverageRelease Notes
Discover more about what's new at AWS with Minimax M2.5 and GLM 5 models now available on Amazon Bedrock
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
Model details
GLM 5
GLM-5 is a large language model from Z.ai (zai-org / Zhipu AI) aimed at complex systems engineering and long-horizon agentic tasks, marking a clear shift from earlier chat-oriented use cases toward multi-step software and tooling work. Compared with GLM-4.5, it scales to 744B total parameters with 40B active at inference, and pre-training data grows from 23T to 28.5T tokens, giving the model a much broader base of knowledge to draw on for extended reasoning and code generation.
A defining architectural change in GLM-5 is the integration of DeepSeek Sparse Attention (DSA), which the creators describe as significantly reducing deployment cost while preserving long-context capacity, making the model more practical to serve in production. The accompanying technical report (arXiv 2602.15763), HuggingFace repository (zai-org/GLM-5), and GitHub repository (zai-org/GLM-5) all point to open-weights distribution, so teams can self-host and fine-tune the model for their own agentic and coding pipelines.
DigitalOceanglm-5glm
Quick Info
Powered by- Provider
- DigitalOcean
- Model key
- glm-5
- Release date
- Feb 11, 2026
- Last updated
- Apr 16, 2026
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $1.00
- Output token cost
- $3.20
Limits
- Output tokens
- 64,000 tokens
- Context window
- 64,000 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.
Latest news about GLM 5
DigitalOcean
CoverageRelease Notes
Chinese AI company Zhipu AI released GLM-5, a 744-billion-parameter open-source model that rivals Claude Opus 4.5 and GPT-5.2 on coding and agent benchmarks.
DigitalOcean
Coverage
The GLM-5 represents a shift in AI development from ‘vibe coding’ to ‘agentic engineering’ to generate an enhanced performance.
DigitalOcean
CoverageBenchmark
LayerLens ran 20 evaluation batches of base GLM-5 across 13 benchmarks on its Stratix harness over a 24-day window spanning February and March 2026. Reported headline scores include 97.4% on MATH-500 and 96.95% on Human Evaluation, placing GLM-5 among the strongest math reasoners LayerLens had tested. On MATH-500 promp The LayerLens re-evaluation surfaced notable non-monotonic behavior: a 12-point regression on Humanity's Last Exam (from 22.4% down to 10.4%) and a 6.66-point improvement on AIME 2025 (from 86.7% up to 93.3%) within the same 24-day window. Failure-mode analysis pointed to arithmetic errors in word problems, constraint