Google DeepMind Cuts Gemma 4 Memory with QAT
Google DeepMind releases Gemma 4 QAT checkpoints, reducing on-device memory for mobile AI deployment.
📂 LLM News
Latest articles in LLM News
NVIDIA Debuts Nemotron 3.5 ASR: Real-Time Multilingual Transcription
NVIDIA launches Nemotron 3.5 ASR, a cache-aware 600M-parameter model transcribing 40 language-locales in real time.
DeepSeek V4: 500x Cost Advantage in Math Proofs
DeepSeek V4 achieves record-breaking efficiency in mathematical reasoning, offering a 500-fold cost reduction over compe…
GPT-5.5 Beats DeepSeek in AI Security Test
Security tests reveal GPT-5.5 has highest success rate in finding APK vulnerabilities, while DeepSeek V4 Pro offers the …
AI Agents: Is the Subagent Model Obsolete?
With 1M context windows, subagents may be unnecessary. Learn why monolithic prompts are winning.
USTC's 30B Model Matches Qwen3-235B Efficiency
USTC open-sources agent-driven training, enabling a 30B model to rival Alibaba's 235B parameter giant in long-context ta…
Step 3.7 Flash Leads AI Speed Rankings
StepFun's Step 3.7 Flash tops Artificial Analysis speed charts with 409 tokens/s, redefining LLM performance benchmarks.
AI Prompt Injection: The Art of Manipulating LLMs
From DAN to PUA tactics, attackers evolve prompt injection methods to bypass AI safety rails. Discover the latest securi…
Google Launches Gemma 4 12B: Encoder-Free Multimodal AI
Google releases Gemma 4 12B, a unified multimodal model processing vision and audio directly on consumer hardware withou…
Microsoft MAI-Base-1 MFU: Why It Trails DeepSeek-V3
Analysis of Microsoft's MAI-Base-1 efficiency metrics reveals significant gaps compared to DeepSeek-V3, highlighting cri…
Mistral AI Unveils Massive Context Window Model
Mistral AI launches a new LLM with an extended context window, revolutionizing long-document processing and enterprise d…
Google DeepMind Releases Gemma 4 12B: Native Audio on Laptop
Google DeepMind launches Gemma 4 12B, an encoder-free multimodal model with native audio support running locally on cons…