Qwen 3.8 27B pushed to 3D game limits

A user pushed Qwen 3.8 27B (q4xl) locally with a PI agent to handle a large game design file, using ~120k context plus vision and MTP. The run read ~11M tokens and wrote ~3.2M tokens over about 12 hours, demonstrating local model capability for complex game design tasks.

• Muse Glimmer Praised by Local Model Users2026-09-15
• AI labs still have a data trust problem2026-09-15
• UkisAI releases Swift‑Qwen3.8‑27B2026-09-14
• Perplexity Portable Computer Now on Windows with NVIDIA RTX2026-09-14
• DeepSeek V4.1 Flash tops AA's new benchmark2026-09-14