NEWSFERENCE
TUE · NOV 17 · 2026 · 19:16 AWST
LIVE
$ today --liveF1TodayF2YesterdayF3ArchiveF4About
NEXT SCAN 19:30

SUN · OCT 25 · 2026

81 clusters from 98 articles.

TOP CLUSTER · TIER 2
Nanbeige opens Nanbeige4.3-33B model weights

Nanbeige released the Nanbeige4.3-33B model on Hugging Face for local experimentation. The model is reported to outperform Qwen3.8-27B across several benchmarks and will support smaller models in the lab's planned distillation pipeline.

ACROSS THE WIRE · TIER 3
20 ROWS
#
ENTITY
HEADLINE
SRC
X
TIME
16
MISC
AI-assisted paper cites nonexistent TorrentFreak article
Drama
1
—
23d
17
PHP
AI-assisted attacks amplify PHP vulnerabilities found with AI
Drama
1
—
23d
18
MISC
AI graveyard tracks projects and startups that failed to deliver
Drama
1
—
23d
19
ANTHRP
Anthropic open-sources sandbox-runtime for OS-level agent isolation
1
—
23d
20
MISC
Local model seekers ask for uncensored reverse-engineering alternatives
1
—
23d
21
KRAFTON
KRAFTON releases Raon 1 Mini preview GGUF
1
—
23d
22
LLAMACPP
llama.cpp CUDA patch fuses FFN and GLU matrix multiplications
1
—
23d
23
SALUKI
Underdog Saluki 27B brings Qwen-based tool calling to local inference
1
—
23d
24
QWEN
Lean-QwenFlash runs Qwen3.8-Flash-Next on 16GB VRAM
1
—
23d
25
QWEN
Custom Qwen coder quantization runs in 8 GB of VRAM
1
—
23d
26
MINIMAX
TensorSharp adds day-one support for MiniMax M3.1 Flash
1
—
23d
27
OPENAI
Community hub revises OpenAI's disputed math write-ups
1
—
23d
28
NVDA
Nvidia open-sources Flora for 125x cheaper robot video labeling
1
—
23d
29
ABLITER
AbliteratedOutTheBox ports ablation engine to Strata
1
—
23d
30
LLAMACPP
llama.cpp fork extends CUDA support to legacy Kepler GPUs
1
—
23d
31
DEEPSK
DeepSeek V4.1 Flash gains local inference support across llama.cpp and quantization tools
1
—
23d
32
LLAMACPP
ik_llama.cpp fork updates MoE optimization and feedback request
1
—
23d
33
QWEN
capy-flash v0.7.4 extends two-GPU Qwen3.8 inference engine
1
—
23d
34
MISC
Open-source browser rhythm game uses four local neural networks
1
—
23d
35
HF
Hugging Face model deletion prompts users to back up weights
1
—
23d