Qwen 3.6 27B vs Gemma 4 31B - making Packman game!
Informal comparison of Qwen 3.6 27B and Gemma 4 31B on local Pac-Man gamedev task; Gemma produced faster, more logical output despite lower token throughput.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Informal comparison of Qwen 3.6 27B and Gemma 4 31B on local Pac-Man gamedev task; Gemma produced faster, more logical output despite lower token throughput.
Qwen3.6-27B achieves 218K context window at 50–66 TPS on single RTX 3090 with stable tool calls after PN12 fix.
Study characterizes consistency of emergent misalignment personas across fine-tuning domains in Qwen 2.5 32B, measuring correlation between harmful behavior and self-assessment.
Qwen 3.6 27B/35B models outperform older ~30B alternatives (Qwen Coder, Gemma) on coding and agent tasks.
User reports empirical comparison of Qwen-3.6-27B running locally vs. proprietary cloud models on coding/hard reasoning tasks.
User reports difficulty replicating reported inference speeds (30-100 tok/s) for Qwen 27B on RTX 3090, achieving 10-19 tok/s instead.
Qwen Team released Qwen-Scope, sparse autoencoders mapping internal features across Qwen 3.5 models (2B-35B), enabling surgical model editing and interpretability.
Qwen 3.6-27B quantized model generates SVG images from creative prompts; community benchmark of image generation capabilities.
Sokann releases Qwen3.6-27B 4.256bpw GGUF quantization achieving 50k context on RTX 5070 Ti 16GB VRAM.
Reddit discussion speculating on future larger MoE variants of Gemma-4 and Qwen 3.6 models.
User received Qwen merchandise sticker; no technical or business substance.
Reddit discussion on Qwen 27B's coding capability relative to closed models; anecdotal developer feedback, no new benchmarks or claims.
Developer built local PDF-to-audiobook app using Kokoro 82M TTS, Qwen, and llama.cpp with Tauri 2.0 on M1 Mac.
Qwen2.5-7B and open-source small models achieve ESI triage accuracy via clinical vignettes without fine-tuning.
IK_LLAMA.cpp adds Qwen3.5 MTP support with 50% throughput gain (18-20→30 tok/s) via pipeline parallelism on 27B model.
Qwen releases FlashQLA, linear attention kernels delivering 2–3× forward and 2× backward speedup for on-device agentic inference via TileLang optimization.
Troubleshooting guide for web search accuracy in Qwen 9B/27B/35B using searXNG, Firecrawl, Jina, and agent prompts.
User reports Qwen 3.6 and Gemma 4 performing expert-level work locally on consumer hardware (RTX 3090), replacing skilled human labor.
Qwen 3.6 27B achieves 60 tok/s throughput and 204k context on dual RTX 5060 Ti 16GB with vLLM.
OcularChat, a Qwen2.5-VL MLLM fine-tuned for age-related macular degeneration diagnosis via visual QA on retinal fundus photographs.
Benchmark comparing abliteration techniques across GLM-4.7-Flash (MoE architecture) vs. prior Qwen family tests; evaluates HauhauCS uncensored claims.
Qwen3.6-27B IQ4_XS quantization bloat analysis; reverting llama.cpp commit reduces VRAM from 15.1GB to 14.7GB with 110k context.
Qwen 3.6 27B quantization benchmark: BF16 (69.78% avg accuracy, 54GB RAM) vs Q4_K_M (66.54%, 15GB) vs Q8_0 across HumanEval, HellaSwag, BFCL.
User reports local LLMs (Qwen 27B, Gemma 4 31B) underperform Claude for coding tasks due to poor tool-calling and decision-making.
Qwen 3.6-27B achieves 38.2% on Terminal-Bench 2.0 coding tasks, demonstrating open-weight models viable for real-world agent work.
GPT-5.5 ranks 2nd on Extended NYT Connections Benchmark; Kimi K2.6 leads open-weights; DeepSeek V4 Pro and Qwen 3.6 Max show significant gains.
Benchmark comparison of 2026 4B-class open-weights models (Gemma 4, Qwen 3.5, Granite 4, Nemotron-3, Phi 4-mini) on M3 Pro.
I'd jump on runpod and ssh in to test my workloads, but they don't have it. Would love to know how well this runs, particularly as context approaches a full 256K. Thanks!
Luce DFlash: open-source speculative decoding for Qwen3.6-27B achieves 1.98x throughput on RTX 3090 via GGUF port.
Production case study: integrating small LMs (Gemma 4E2B, Qwen3 0.6B) into Android app reveals memory, latency, and quantization challenges for on-device deployment.