We’re rolling out Gemini 3 Flash starting today. Here’s where you can find it:
LLMS
-
Google Launches Gemini 3 Flash Model with Advanced Reasoning
By
–
We’re expanding the Gemini 3 family with the launch of Gemini 3 Flash. This model:
— Google AI (@GoogleAI) 17 décembre 2025
— Combines Gemini 3’s Pro-grade reasoning with Flash-level latency, efficiency, and cost
— Delivers frontier-level performance on PHD-level reasoning and knowledge benchmarks
— Is our most… pic.twitter.com/s7fM0Z6vRxWe’re expanding the Gemini 3 family with the launch of Gemini 3 Flash. This model: — Combines Gemini 3’s Pro-grade reasoning with Flash-level latency, efficiency, and cost
— Delivers frontier-level performance on PHD-level reasoning and knowledge benchmarks
— Is our most -

Existential Problems in LLM Serving and Autoregressive Inference
By
–
The Existential Problems in LLM Serving Naive Transformers might be fine for lab experiments – but they don’t hold up in production. The real challenge lies in Autoregressive Inference, where performance bottlenecks can cripple even the most powerful GPUs.
If you’ve ever seen -

Improving RAG with Forward and Backward Lookup
By
–
Improving RAG with Forward and Backward Lookup This is a clever use of small and large language models. Traditional RAG systems compute similarity between the query and context chunks, retrieve the highest-scoring chunks, and then generate. But complex queries often lack
-

Hindsight: Agent Memory System for Smarter Learning
By
–
Agent Memory That Works Like Human Memory! Hindsight is an agent memory system built to create smarter agents that actually learn over time. Here's the biggest problem: Existing open-source memory solutions rely heavily on RAG, vector databases, and knowledge graphs. These are
-
Chinese AI Companies Pivot to Open Source Over AGI Competition
By
–
It’s due to LLM/AGI. The opportunity has gone from within-China to global. And while top American LLM companies chose to pursue AGI, Chinese companies cannot raise the funds to compete, so they’ve chosen open source
-

LLaVA-UHD v3 Halves Processing Time With Progressive Visual Compression
By
–
Ever wonder why AI vision models are so slow? This research shows the quest for high-resolution image understanding creates a major speed bottleneck. The solution, LLaVA-UHD v3, uses a clever "Progressive Visual Compression" method to cut processing time by over half while
-

Mistral Announces New Creative Model
By
–
Mistral announced a new mistral-small-creative 32k model on their Discord! An experimental small model designed for creative writing, roleplay and character-driven dialogue, general-purpose instruction following, and conversational agents.
-

Foundation Model Transparency Index 2025 Shows Sharp Decline
By
–
The 2025 Foundation Model Transparency Index is here! This report reveals a sharp drop in openness, with average scores plummeting from 58 to 40. While IBM leads with 95, xAI and Midjourney scored just 14. – Google: 41
– OpenAI: 35
– DeepSeek: 32
– Qwen: 26 The 2025 -

Error-Free Linear Attention Surpasses Hybrid Attention Methods
By
–
With the recent hybrid attention releases from MiniMax, Qwen, Kimi, and NVIDIA, this paper introduces Error-Free Linear Attention that could top them all This new technique has a stable linear-time attention that's better than any linear attention variants and also DeltaNet!
