GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
xupy21
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
xupy21
Model Tree
Base
Qwen/Qwen3-VL-8B-Instruct
Fine-tuned
this model
Input Modalities
Text
Image
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
Container
Across 12 diverse multimodal benchmarks, ContextRL improves over the standard GRPO baseline by +1.6 points on average, while improving every individual benchmark.
| Benchmark | Base | RL (GRPO) | ContextRL (Ours) |
|---|---|---|---|
| MathVista | 75.8 | 78.7 | 79.8 |
| MathVerse | 56.1 | 65.0 | 66.4 |
| MathVision | 46.2 | 49.2 | 52.0 |
| MMMU-Pro | 41.3 | 55.9 | 57.5 |
| MMMU | 66.4 | 69.1 | 70.1 |
| V* | 82.2 | 84.8 | 85.9 |
| MMStar | 70.5 | 73.5 | 74.8 |
| BLINK | 64.4 | 65.1 | 66.6 |
| ScienceQA | 94.4 | 95.6 | 96.6 |
| PhyX | 45.5 | 72.1 | 73.4 |
| OlympiadBench Phy | 7.9 | 8.1 | 9.9 |
| MME-RealWorld Lite | 48.7 | 51.9 | 54.8 |
| Overall Avg. | 58.3 | 64.1 | 65.7 |
This model follows the same interface as Qwen3-VL-8B-Instruct and can be loaded with
transformers. Training and evaluation code, data construction pipelines, and detailed
configurations are available in the repository:
👉 https://github.com/xupy2003/ContextAwareRL
Please refer to the repo's README for environment setup, inference scripts, and reproduction instructions.