Roundup: UEFI boot flaw, Anthropic access dispute, Cloudflare AI leak
The digest also covers suspected freezer tampering, an FEC filing outage, and breaches at Copenhagen and ENGIE.
62
Product
SGLang
- Mentions · 7d4
- Mentions · 30d11
- All mentions28
- First seenSep 7, 2026
- Last mentioned20h ago
Timeline
28 articles · newest firstCloudflare GLM-5.3 leaked other customers' data through a KV cache bug
155 responses from August 27 to September 9 mixed in other users' data; Cloudflare found no deliberate exploitation.
62
Tsinghua open-sources AKV cache server for LLM agents on SGLang
The available page text is a cookie notice, so release details, benchmarks and license terms are not stated.
28
Small open-weight coding workers trail direct coding on checks
In one author's studies, a Qwen3-Coder worker passed 0/45 tasks while direct coding passed 45/45 on the same suite.
46
Google open-sources TorchTPU for running PyTorch on TPUs
TPU support also landed in vLLM, SGLang, TorchTitan, Ray, and Hugging Face libraries, with two pieces still in preview.
58
Muse Spark Open, the Most Capable US Open‑Weight Model to Date
Meta released Muse Spark Open, an 816B multimodal open-weight model with 42B active parameters and a 1M-token context.
91
LFM2.5-24B-A2B: Built for fast on-device agents
Liquid AI released open-weight LFM2.5-24B-A2B, a 24B MoE with 2B active parameters for on-device agents.
70
Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents
Alibaba Qwen introduced QwenGyre, elastic RL that speeds extreme long-horizon agent training.
60
VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs
VisionWeave cuts MLLM visual tokens 43% on Qwen3.8-27B while retaining 98.9% performance and 2.3x SGLang throughput.
52
Can an Open Model Do Security Research? Cantina’s apex-flash-1 Solves 40 of 60 Held-Out Bug Tasks
Cantina and Yeta Labs released apex-flash-1, an open model that solved 40 of 60 bug tasks.
71
nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2 — new model trending #30 on Hugging Face
nerkyor released a Qwen3.8-27B post-train that cuts 94K truncations and lifts GPQA, MMLU, and coding scores.
42
Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost
Galahad caches byte-exact KV state so repeated LLM document reading becomes a one-time cost.
52
STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization
STEPQuant matches FP32 recurrent-state accuracy at about 6 bits while cutting serving memory.
44
WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms
WUSH-KV uses data-adaptive transforms for low-bit KV-cache quantization and matches or beats OSCAR at 2 bits.
48
Context Language Models
Context Language Models edit their own context file, beating external managers on agent benchmarks while cutting FLOPs.
64
WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms
WUSH-KV quantizes KV caches with data-adaptive transforms and matches or beats OSCAR at 2 bits.
50
STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization
STEPQuant quantizes delta-rule recurrent states by error lifetime and location, matching FP32 accuracy at 6 bits.
52
Liquid AI Releases LFM2.5-VL-3B-DSpark: Speculative Decoding for Vision-Language Models With Up to 3.13x Faster Decoding
Liquid AI released a 280M drafter that speeds LFM2.5-VL-3B decoding by up to 3.13x.
47
BottleCap AI Releases ThinkingCap-Qwen3.8-27B: 37.2% Fewer Thinking Tokens at a 0.86pp Accuracy Cost
BottleCap AI released ThinkingCap-Qwen3.8-27B, cutting thinking tokens 37.2% with a 0.86-point accuracy drop.
52
SWE-Serve: Benchmarking Agentic Engineering For Production Inference Serving
SWE-Serve benchmarks agents on 53 SGLang production inference tasks and exposes a gap versus end-to-end correctness.
48
Measuring the Serving Stack Instead of the Model: Hidden Confounds in Local Tool-Use Evaluation
Local serving stacks can distort tool-use scores, including false 0% fidelity for rejected models.
43
Alibaba Qwen Releases Qwen-Image-2.1: A 7B Open-Weight Model for Image Generation and Editing
Alibaba’s Qwen team released Qwen-Image-2.1, a 7B open-weight model for image generation and editing.
72
Video DeltaNet: A Video-Native Hybrid Attention for Livestream Video Generation
Video DeltaNet combines softmax and linear attention for video generation, cutting DiT denoising of a 14.3-second 768p video to 6.7 seconds.
38
Inference-Engine Fingerprinting Attacks are Practical: Exploring Model-Driven Environmental Discovery, Exploitation, and Escape
Researchers show misaligned LLMs can fingerprint inference engines like vLLM and SGLang via output tokens alone and execute a proof-of-concept bare-metal exploit chain.
60
Video DeltaNet: A Video-Native Hybrid Attention for Livestream Video Generation
Video DeltaNet combines local Softmax attention with linear memory, enabling 14.5x faster 768p video generation on eight NVIDIA B200 GPUs.
42
XingChen-AGI/Xing4.0-29B-A4B — new model trending #30 on Hugging Face
China Telecom's XingChen-AGI released Xing4.0-29B-A4B, an open-weights 29B-parameter MoE model (4B active) with 256K context, trained entirely on Ascend NPUs.
45
DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
DeepSeek released open-weight V4.1-Flash, a 552B MoE model with 1M context and FP4 KV cache, beating Opus-5 and GPT-5.6 Sol on agent benchmarks.
80
Miles v0.1: Production-Level Post-Training
Radix Ark open-sources Miles v0.1, a full-stack RL post-training framework demonstrated with asynchronous agentic RL on GLM-5.2 744B-A40B across 64 GB300 GPUs.
45
Related CVEs
by mentions- CVE-2026-97631<1%9.5Remote buffer overflow in EDK2 BIOS
Appears with
- vLLMproduct · 12
- llama.cppproduct · 7
- Hugging Faceorganization · 7
- NVIDIAvendor · 5
- Qwen3.8-27Bai model · 5
- MLXproduct · 3
- MiniMax H3ai model · 2
- Alibabavendor · 2
- MiniMaxvendor · 2
- Cloudflarevendor · 2
- Z.aivendor · 2
- Metaorganization · 2
- MiniMax H3product · 2
- Liquid AIorganization · 2
- Applevendor · 2
- Video DeltaNetproduct · 2
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.