Cited source trail
Local model landscape source trail - July 3, 2026
Original question
GLM-5.2 (Zhipu AI) 744B MoE model info: visit huggingface.co/chat/models find GLM-5.2 model card, BenchLM leaderboard page, github.com/THUDM/GLM for official repo, license info, hardware requirements. Also check Qwen 3.6 model card on Hugging Face, DeepSeek V4-Flash model card. For each: cite the official model card URL, Hugging Face page, GitHub repo, license type, parameter count, and benchmark scores from BenchLM or independent sources. Provide direct clickable links to every source.
Sources checkedhuggingface.co
Next step
Keep going with a follow-up, or copy the result if you want to share it.
Below is the requested information for the specified models as of July 2026.
### 1. GLM-5.2 (Zhipu AI)
* **Model Card URL:** [huggingface.co/zai-org/GLM-5](https://huggingface.co/zai-org/GLM-5) (Note: GLM-5.2 is the current iteration hosted here).
* **GitHub Repo:** [github.com/THUDM/GLM](https://github.com/THUDM/GLM)
* **License:** MIT
* **Parameter Count:** 744B total parameters (~40B active, MoE architecture).
* **Benchmark Scores (Selected):**
* **HLE:** 30.5
* **GPQA-Diamond:** 86.0
* **SWE-bench Verified:** 77.8
* **τ²-Bench:** 89.7
* **Hardware Requirements:** High-end enterprise GPUs (e.g., H100/H20) are recommended for full deployment; optimized inference via vLLM, SGLang, or KTransformers is required for efficient local hosting.
### 2. Qwen 3.6 (Alibaba Cloud)
* **Model Card URL:** [huggingface.co/Qwen/Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B)
* **GitHub Repo:** [github.com/QwenLM/Qwen3.6](https://github.com/QwenLM/Qwen3.6)
* **License:** Apache 2.0 (standard for Qwen series)
* **Parameter Count:** 35B total (3B active) for the MoE variant; 27B for the dense variant.
* **Benchmark Scores:** Qwen 3.6-35B-A3B is noted for flagship-level agentic coding performance, surpassing Qwen 3.5-35B. Detailed leaderboard rankings are available on [benchlm.ai](https://benchlm.ai).
### 3. DeepSeek V4-Flash
* **Model Card URL:** [huggingface.co/deepseek-ai/DeepSeek-V4-Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash)
* **GitHub Repo:** [github.com/deepseek-ai/DeepSeek-V4](https://github.com/deepseek-ai/DeepSeek-V4)
* **License:** MIT
* **Parameter Count:** 284B total parameters (13B active, MoE architecture).
* **Benchmark Scores:** Positioned as a high-efficiency reasoning model. Independent rankings and comparative scores are tracked on [benchlm.ai](https://benchlm.ai).
### Resources
* **BenchLM Leaderboard:** [benchlm.ai](https://benchlm.ai) (The primary source for independent, cross-model benchmark comparisons).
Sources visited
1searchGLM-5.2 Zhipu AI 744B MoE model
2searchQwen 3.6 model card Hugging Face
3searchDeepSeek V4-Flash model card
4searchBenchLM leaderboard
5navigatehttps://huggingface.co/zai-org/GLM-5
Shared by Unchained Sky · Jul 4, 2026
Ask a follow-up
Refine the angle, compare sources, or ask for a deeper cut on this answer.