GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
AxionML
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
AxionML
Model Tree
Qwen/Qwen3.6-35B-A3B
Input Modalities
Output Modalities
Supported Functionality
| Architecture | Mixture-of-Experts with hybrid attention (Gated DeltaNet + Gated Attention) |
| Total Parameters | 35B |
| Activated Parameters | 3B |
| Input | Text, image, video |
| Context Length | 262K tokens |
| Checkpoint Size | ~23.4 GB |
| Benchmark | BF16 | NVFP4 |
|---|---|---|
| MMLU Pro | 85.6 | 85.0 |
| GPQA Diamond | 84.9 | 84.8 |
| τ²-Bench Telecom | 95.5 | 94.7 |
| SciCode | 40.8 | 40.6 |
| AIME 2025 | 89.2 | 88.8 |
| AA-LCR | 62.0 | 62.0 |
| IFBench | 62.3 | 62.8 |
| MMMU Pro |
Scores reported by NVIDIA for this checkpoint. Baseline: Qwen/Qwen3.6-35B-A3B. SciCode with
temperature=0.6; otherstemperature=1.0;top_p=0.95, max 131,072 tokens.
bash
python3 -m sglang.launch_server \ --model-path AxionML/Qwen3.6-35B-A3B-NVFP4 \ --quantization modelopt_fp4 \ --tp 1 \ --reasoning-parser qwen3bash
vllm serve AxionML/Qwen3.6-35B-A3B-NVFP4 \ --quantization modelopt \ --max-model-len 262144 \ --reasoning-parser qwen3Upstream validation was done with vLLM (vllm/vllm-openai:nightly) on GB300.
The base model was trained on data that may contain toxic language and societal biases. The quantized model inherits these limitations. It may generate inaccurate, biased, or offensive content. Please refer to the original model card and the upstream quantized model card for full details.
| 74.1 |
| 74.5 |