Qwen3 235B-A22B — In-Depth LOCAL Testing! (The BIGGEST Qwen Yet!)
AI Summary
Summary of YouTube Video: Quen 3235B-A22B Model Overview
Overview
- Introduction to the Quen 3235B-A22B model, released recently.
- Comparison to top-tier models: Deepseek R1, 0103 Mini, Gro 3, Gemini 2.5 Pro.
System Specifications Used
- Equipped with:
- 2 x 3090 Ti GPUs
- 128 GB DDR4 RAM
- Intel i7 (12000 series)
- MSI motherboard.
- Context length set at 4,096.
- GPU offload at 30 out of 94.
- Uses approximately 90 GB of system RAM.
- Response speed: 2.5 to 3 tokens per second.
Key Performances and Tests
- HTML Website Test
- Generated a simple PC repair website (StevesPCRepair) with basic features and testimonials.
- Mixed review on layout and aesthetics; some elements were well-executed, while others were lacking.
- Game Development
- Created a retro synth wave obstacle avoidance game titled “Neon Drift.”
- Initially presented issues with graphical elements but functioned correctly overall.
- Adapted the game to a Pong variant, indicating flexibility in design.
- Website Generation Enhancements
- Requested a more impressive version of the PC repair website, resulting in a cleaner layout but still not fully satisfying.
- VC Pitch Generation
- Developed a mock SaaS product concept named “Notify” focused on automating notifications in workflows.
- Feedback emphasized the feasibility of presented ideas but raised concerns about the authenticity of generated links.
Conclusion
- While the Quen model demonstrated solid capabilities, it exhibited limitations compared to state-of-the-art models due to quantization.
- The results were functional and educational, highlighting both strengths and areas for improvement.
- Future plans include exploring agentic uses and enhanced coding IDE integrations.