New AI startup could shrink server-sized models for use on iPhones
…The company has already been able to compress the 54GB Qwen 3.6 model to just 4GB. And, importantly, the technology that it uses doesn't impact the model's performance. Apple…
Search
All sources
huggingface.co
50
xda-developers.com
14
blogs.nvidia.com
11
developer.nvidia.com
7
amd.com
7
wccftech.com
6
neowin.net
4
aws.amazon.com
3
intel.com
2
cnx-software.com
2
theverge.com
2
heise.de
2
Tracked topic
Qwen3
Qwen3 is an AI model family developed by Alibaba, released as a set of large language models for natural-language tasks.
Live now
Local one-shots, you say? Yep, we're there with Qwen 3.8 27B
Top stories
Paper page - KVpop -- Key-Value Cache Compression with Predictive Online Pruning
…Lukas Hauzenberger , , , , , , , Abstract KVpop learns optimal key-value cache eviction by directly supervising keep-or-drop decisions using future-attention targets, achieving high performance with reduced memory usage. Generated by Qwen/Qwen2…
Alibaba Targets NVIDIA's Hopper With Zhenwu M890 AI Chip, Claiming 3x The H20 Performance, 144GB HBM3 & A Roadmap Through 2028
…Notably, Qwen 3.7-Max can autonomously execute long-horizon agentic tasks—sustaining continuous operation for up to 35 hours and managing over 1,000 tool calls without performance degradation. Deeply optimized…
Lokale KI schneller machen: So zünden Sie den LLM-Turbo
…auf einer Radeon RX 9060 XT für Gemma 4, Qwen3.5 und Qwen3.6 sowie auf einer Nvidia GeForce RTX 3090 für Gemma 4, Qwen3.6 und Qwen3.8. Die Geschwindigkeitsgewinne sind…
Discussions and forums
Tesla V100 Qwen3.6 27B Performance
Looking for V100 users to share your config and it's performance. GPU: Tesla V100 PCIE 32Gb Qwen3.6 27B Q4_K_M + Q8_0 MTP 128K context length Pi coding agent llama.cpp model preset: [*] spec-default = 1 ctx-size = 131072…
Qwen3.8-Max matches Kimi K3 and DeepSeek V4 Flash
Qwen3.8-Max (2.4T) is another massive contribution to the open weight community. On benchmarks, it performs closely to Kimi K3 and DeepSeek V4 flash across all categories and is better at coding and software tasks. Qwen3…
I'm really hoping we're in 2026's 2-month-gap between QwQ and Qwen3 right now
QwQ was genuine next-gen performance usable on local hardware, but the massive required context (it's reasoning style was akin to "if I say every possible word, I'll notice the right one!") kinda made it unusable for age…
Ultrafast Qwen3-TTS at 34 ms Time-to-First-Audio, Handling 10 Requests Per Second [OSS]
Hey locallama! We recently open sourced a Qwen3-TTS 1.7B implementation that achieves 10 requests per second (RPS) and sub-50 ms p95 time-to-first-audio (TTFA) while maintaining real-time playback on 1 x H100. This exten…
Medical model: Reasoning-Medical-27B (Qwen3.6-27B finetune)
From the description: "Reasoning-Medical-27B is designed for universal advanced medical reasoning in professional medicine, medical genetics, college biology/medicine, and clinical knowledge. The model was fine-tuned on …
SageMaker AI now supports serverless model customization for Qwen3.6 - AWS
SageMaker AI now supports serverless model customization for Qwen3.6 Posted on: May 14, 2026 Amazon SageMaker AI now supports serverless model customization for Qwen3.6 27B parameter model using supervised fine…
Paper page - Little Brains, Big Feats: Exploring Compact Language Models
…Qwen-family models dominate the top-performing SLM tier in this setting. Qwen3-8B-Q4_K_M achieved the strongest overall SLM quality, reaching 0.72 correctness and 0.83 faithfulness, approaching…
Paper page - KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking
…On BEIR , KaLM- Reranker -V1 achieves state-of-the-art performance, on par with strong industrial models such as the Qwen3- Reranker series; on MIRACL , despite not being extensively trained on multilingual…
Alibaba delivers RISC-V server chip optimized for Chinese AI
…Qwen3 and DeepSeek V3, potentially becoming a new type of high-end CPU for the AI Agent era,” the post enthuses. Alibaba claims the machine’s single-core general-purpose performance “exceeded…
DeepSeek's Peak-Hour Pricing Betrays Where Its Users Really Live, Calming US Fears of a China AI Takeover, Even As Alibaba's Qwen Models Bury Meta On Hugging Face
…First up, the Qwen 3.8-Max (2.8 trillion parameters) is second only to Moonshot's Kimi K3 in terms of performance. More intriguing, however, is the Qwen 3.8 (27…
Paper page - From Trainee to Trainer: LLM-Designed Training Environment for RL with Multi-Agent Reasoning
…With Qwen3-4B as the backbone, our framework achieves the strongest aggregate performance on our benchmarks, outperforming larger proprietary LLMs (e.g., GPT, Gemini) and fixed-environment training baselines. We further analyze…