AI Briefing#qwen
The latest AI and developer news about #qwen, with the original source and a short summary.
The latest AI and developer news about #qwen, with the original source and a short summary.
The 9B-scale Qwopus3.5-9B-Coder model, optimized for agentic coding and tool calling, has been released.
A template error in the Qwen3.5 Vision model has been fixed in vLLM.
Hermes Agent was introduced as a self-improving local AI agent for NVIDIA RTX PCs and DGX Spark.
Fine-tuning Qwen3.5-4B with RL for RLM achieved performance on par with Claude Sonnet 4.6.
Luce DFlash/PFlash boosts LLM inference performance by up to more than 3x on the AMD Strix Halo iGPU.
A local 35B model handled about half of daily tasks faster than the cloud.
By applying a caching technique that computes long prompts only once during RL training, training efficiency was significantly improved.
Unsloth has released GGUF versions of Qwen models with Multi-Token Prediction (MTP) technology applied.
MTP support in llama.cpp can boost inference speed for the Qwen 3.6 27B model by 2.5x.
MTPLX, an engine that leverages MTP heads to boost LLM inference speed by up to 2.24x on Apple Silicon, has been released.
vLLM has resolved the unimplemented TurboQuant error affecting Qwen 3.5+ models caused by an issue with Mamba layers.
APEX, a mixed-precision quantization strategy for MoE models, has expanded to over 30 model types and launched a new ultra-compressed tier.
A reproducible setup was shared for running Qwen 3.6 35B MoE on an RTX 3090 with 262K context.
Ran Qwen3.6-35B-A3B at 23 t/s on a 5-year-old laptop with 6GB VRAM.
The latest llama.cpp build significantly improved the RTX6k Qwen3.5 benchmark.
Qwen has released a Top-K SAE interpretability module and checkpoints for 64 layers for Qwen3.5-27B.
CAJAL-4B-P2PCLAW has released a 4.21B parameter model specialized for distributed systems.
Qwen3.6-27B achieved 95.7% on SimpleQA in a single RTX 3090 local environment.
On Windows 10 with native vLLM, Qwen3.6-27B hit 72 tok/s.
The RTX 6000 Pro was significantly faster than the 4080 Super at Qwen3.6 27B inference.
RedHatAI released the Qwen3.6-27B FP8 model on Hugging Face.
DFlash for Qwen3.5-35B-A3B was confirmed running on an RTX 2080 SUPER 8GB.
In a MacBook Pro M5 Max test, Gemma 4 31B outperformed in Pac-Man generation quality.