RF-DETR, Roboflow's SOTA real-time detection and segmentation model beating YOLO, is now part of HuggingFace Transformers with low-VRAM fine-tuning and a live webcam demo.
HuggingFace releases LeRobot Humanoid — a fully open-source bipedal robot platform at ~$2,500 build cost with 3D-printed hardware, runtime, and training environments included.
Dell and Hugging Face launch dell.hf.co, enabling one-click on-premises deployment of Kimi K2.6, DeepSeek V4 Pro, GLM 5.1, MiniMax M2.7, and DeepSeek V4 Flash on PowerEdge XE9780+B300.
Hugging Face releases Ettin Reranker: 6 CrossEncoder models from 17M to 1B parameters, SOTA at every size, built on ModernBERT with a publicly released training recipe.
Andrej Karpathy confirms joining Anthropic for frontier LLM R&D, calling the next few years 'especially formative' — and open-source implications loom large.
HuggingFace skills let Claude Code fine-tune a VLM with a single prompt: agent calculates VRAM, picks instance, launches job on HF infra. Fully agentic model training.
Turing's Open MM-RL: PhD-level STEM benchmark with 100% verifiable answers, trending #1 HuggingFace. Every prompt double-vetted by PhD specialists. 3,000 more tasks coming.
Qwopus3.6-35B: Qwen3.6 35B distilled on Claude Opus reasoning traces (71.9GB) brings Opus-quality reasoning patterns to an open-weight architecture.
Endless Terminals — an autonomous RL task generation pipeline with zero human annotation — hit 73k+ Hugging Face downloads in its first month with TerminalBench 2.0 gains.
Hugging Face and Pollen Robotics launched a 300+ app open-source App Store for the Reachy Mini robot, with 10,000 units already deployed globally.
Turkey's Presidential Communications Directorate joins Hugging Face as first public institution on the platform — HF CEO Clément Delangue calls for global government sovereign AI via open-source.
HuggingFace's ml-intern autonomously trained nanowhale (100M MoE), pushed GPQA from 10% to 32% in 10h, and beat Codex on HealthBench by 60%. CLI + web app now public.
Agent-generated PRs to HuggingFace transformers quadrupled; bulk-merging hundreds of them showed zero benchmark regression on arc_challenge, gsm8k, hellaswag.
AgentTrove is a new 1.7M-sample agentic training and evaluation dataset released by OpenThoughts on HuggingFace.
APEX-Agents benchmark for consultant/lawyer/banker-level AI work now has a Hugging Face leaderboard for open-source model evaluation.
TIME100 names Hugging Face (top-10 AI) and Mistral AI among 2026's most influential companies—both positioned explicitly against closed-model lock-in.
huggingface_hub v1.13.0 adds --format, --json, and -q global flags to every hf command, with auto mode adapting output for humans vs. AI agents.
HuggingFace releases smol-audio: an open notebook collection covering local fine-tuning of Whisper, Parakeet, Voxtral, Audio Flamingo 3, and Dia-1.6B TTS — a complete local audio AI toolkit.
Unsloth, the fine-tuning optimization library, has surpassed Microsoft on HuggingFace to enter the top 10 most-followed organizations — reflecting open-source momentum over Big Tech on the platform.
Hugging Face forms a dedicated PyTorch/MPS team targeting 100× Apple Silicon perf gains — torch.sort and torch.multinomial are already MPS-native; flex attention is next.
DeepSeek V4-Pro open-sourced with 1.6T params, 1M context window, and 10x KV cache reduction vs V3.2 — #1 HuggingFace trending in 43 minutes.
HuggingFace crosses 1.2 million hosted AI apps, positioning it as likely the world's largest AI app store by application count.
DeepSeek V4-Pro launches with 1.6T parameters, 1M context, and 10× KV cache reduction over V3.2 — multiplying inference concurrency roughly 10× on the same hardware.
HuggingFace Inference Providers undercuts OpenRouter with zero markup on 200+ models, surfacing as the preferred open-model routing layer for cost-sensitive deployments.
HuggingFace's ml-intern autonomously runs the full ML research-to-training loop, lifting GPQA from 10% to 32% on a 1.7B model in <10 hours and beating Codex on HealthBench by 60%.
ml-intern reads arXiv, cleans datasets, runs SFT/GRPO, diagnoses failures, and iterates — pushing GPQA from 10% to 32% in under 10 hours for roughly $1 of compute.
Curated AI insights — sent when there's something worth your inbox.