Liquid AI's 300M Draft Model Speeds Up Decoding by Up to 3.18x
DSpark released for three LFM2.5 models — greedy output identical to the original, MacBook throughput up 61→139 tokens/sec
10
Hugging Face
A public marketplace where AI models are gathered and distributed. Rather than a company that builds models itself, it functions as "the GitHub of models," hosting publicly available models built by others. The Hub hosts a wide range of open-weight models, from transformer-based models to image and speech generation models. The structure is such that various providers plug in inference infrastructure for users who want to download and run models themselves, and in August 2026, Baseten joined this list as an inference provider for Hugging Face.
Current rank (1M)
6
- Change
- Prev.
- 3
Rank over the last 30 days
2026.08.17 – 2026.09.15 · High 3 · Low 6 · Now 6
Hugging Face · Related stories
Cartesia Sonic-3.6 tops both voice synthesis leaderboards
It won even in a test that standardizes voices and pits only the engines against each other, using an SSM architecture instead of a transformer
90
AI agent memory needs different prescriptions by model size to boost performance
In IBM Research's tests across 8 models, DeepSeek preferred full guideline sets while gpt-oss performed better with compressed versions
00
Claude and Codex build toy tokenizers in 30 minutes, then fail at production scale
Liquid AI added a loop, and both coding agents completed the production-grade task
40
OpenAI tightens monitoring and isolation after Hugging Face incident
New monitoring aims to flag alerts within 30 minutes, but largest frontier RL training remains paused
20
Litter-Robot 5 Pro: The AI Mistook Two Cats for One
Six months with the $899 AI robot litter box — waste removal was excellent, but facial recognition and waste detection fell short
50
MiniMax unveils music model that generates full 5-minute songs from lyrics alone
Open-weight MiniMax-Music3 generates a complete 32kHz stereo track in one pass from just two inputs: lyrics and a structured caption
30
Hugging Face breach by unreleased OpenAI model prompts expert warning
Former OpenAI staffer Brundage tells Bloomberg third-party audits are needed
20
Open-source app lets you grow a mini LLM from scratch on a MacBook
LabLLM supports GPT-style model design, training, and chat locally on Apple Silicon via MLX
30
Artificial Analysis launches Optima, a tool for benchmarking AI models on your own data
New custom benchmarking platform compares not just quality but cost and time per task
20
Qwen3.8-27B tops Hugging Face trending models
The open-weight model, released on the 14th, topped the charts in downloads and likes
20
Qwen confirmed as No. 1 in Hugging Face's local inference rankings
Hugging Face's "State of Open Models" report shows Qwen leading Gemma in real-world local inference use
10
Hugging Face · Recent announcements
- 470StudentSim: Training LLM-based Student Simulators
- 346Qwen-Drive-1.0: An Initial Step towards a Vision-Language Foundation Model for Autonomous Driving
- 185Agentic Game Development as a Verifiable Trajectory Data Engine for Scaling World Models
- 169VoiceMem: Streaming Dual-Brain Memory for Real-Time Interaction
- 137PAWBench: How Far Are We from Probabilistically Aligned World Modeling?
- 7,881🤗💚
- 105UrbanGround: From Local Perception to Spatial Agency in a Real-Scale City
- 98Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
- 93DreamX-Creator: Democratizing Native Audio-Video Generation at 2K Resolution
- 94LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
- 77SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers
- 72Lucida: Parse, Generate, and Place for Composable Real-to-Sim Scene Modeling
- 72Self-OPD: On-Policy Distillation for Flow Matching Models without Teacher
- 73TTPO: Test-Time Policy Optimization
- 68Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models
- 63What Makes Good Agentic Data? An ACE Lens on Data Generation for LLM Agents
- 59GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling
- 56Agentic Artifact Creation: Systems, Evaluation, Principles, and Opportunities
- 56UI-Venus-2 Technical Report
- 47VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning
- 42H3-World: Turning Language Understanding into World Control
- 43Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning
- 40ZimaBlue: Evolving Generalizable World Action Models through Scalable Video Pre-training
- 39Normalized Low-Rank Adaptation
- 35PaperGym: Rubric-Centered Evolution for Research-Plan Generation