Qwen 27B long-context test: 500K tokens with fp16 KV cache holds up
Constant_Art_20 · reddit · 2026-10-30
The author tested Qwen 3.8 27B by having it build and animate a DeepSeek-style chibi whale in Blender, probing rope stability. Result: 500K context with fp16 KV cache is viable, and the model navigated new tools on its own for the chibi-style pipeline.
More from Models
- Xiaomi open-sources Xiaomi-OCR-0, a 0.8B document understanding model topping sub-1B parsing — gnukeith · 2026-10-30
- Unsloth says it can't reproduce GSQ-RCO results, ships DeepSeek-V4.1 Dynamic v3.1 GGUFs — danielhanchen · 2026-10-30
- OpenAI's subscription pain: low limits and unbanked resets drag utilization — teortaxesTex · 2026-10-30
- Subscription utilization test: Claude Max hits 80-90%, OpenAI only ~50% — teortaxesTex · 2026-10-30
- Unsloth rebuts GSQ-RCO: DeepSeek-V4.1-Flash 1bit AIME25 is 91.88%, not 100% — danielhanchen · 2026-10-30
- SAM3 now runs offline on your GPU: ONNX export hits 0.37s on M5, 0.08s on RTX 3080 Ti — CSProfKGD · 2026-10-30