AI Pulse by Inblix

AI Research

Cutting-edge AI research papers, breakthroughs, and academic developments.

587 articles

NVIDIA's Cosmos Reason 2 tops physical AI charts, giving robots better common sense — Inblix summary
Research

NVIDIA's Cosmos Reason 2 tops physical AI charts, giving robots better common sense

Hugging Face Blog · Jan 5, 2026 · 2 min read

Robots are about to get a lot less clumsy. NVIDIA just dropped Cosmos Reason 2, an open vision-language model that imme...

Falcon-H1-Arabic Hits 256K Context, Outperforms Bigger Models with Hybrid Design — Inblix summary
Research

Falcon-H1-Arabic Hits 256K Context, Outperforms Bigger Models with Hybrid Design

Hugging Face Blog · Jan 5, 2026 · 2 min read

The team behind Falcon-Arabic just dropped a sequel that isn't a minor tune-up — it's a ground-up rebuild. Falcon-H1-Ar...

NVIDIA just showed how to build your own desk buddy robot with DGX Spark — Inblix summary
Research

NVIDIA just showed how to build your own desk buddy robot with DGX Spark

Hugging Face Blog · Jan 5, 2026 · 2 min read

Jensen Huang walked onto the CES 2026 stage and didn't just talk about AI agents — he showed one you could build yourse...

ServiceNow drops an 8B-param guard model built to spot attacks that regex filters miss — Inblix summary
Research

ServiceNow drops an 8B-param guard model built to spot attacks that regex filters miss

Hugging Face Blog · Dec 23, 2025 · 2 min read

ServiceNow's research team just open-sourced AprielGuard, an 8 billion-parameter model designed to be the safety classi...

Hugging Face's v5 tokenizer overhaul separates architecture from vocab for custom training — Inblix summary
Research

Hugging Face's v5 tokenizer overhaul separates architecture from vocab for custom training

Hugging Face Blog · Dec 18, 2025 · 2 min read

Hugging Face just made tokenization less of a black box. The Transformers v5 release fundamentally redesigns how tokeni...

NVIDIA dares the industry: reproduce our Nemotron Nano 3 scores yourself — Inblix summary
Research

NVIDIA dares the industry: reproduce our Nemotron Nano 3 scores yourself

Hugging Face Blog · Dec 17, 2025 · 2 min read

Most model benchmarks are marketing theater. NVIDIA is betting that showing its work changes the game. Alongside the Ne...

CUGA agent snags #1 on AppWorld benchmark, runs 90% cheaper on open models — Inblix summary
Research

CUGA agent snags #1 on AppWorld benchmark, runs 90% cheaper on open models

Hugging Face Blog · Dec 15, 2025 · 2 min read

There’s a new open-source agent topping the leaderboards, and it’s designed to make building AI coworkers less of a hea...

llama.cpp now juggles multiple models without crashing, just like Ollama — Inblix summary
Research

llama.cpp now juggles multiple models without crashing, just like Ollama

Hugging Face Blog · Dec 11, 2025 · 2 min read

The llama.cpp project just shipped one of its most-requested features: native model management that lets a single serve...

OpenAI's Codex Can Now Train Its Own Open-Source Models, But Should It? — Inblix summary
Research

OpenAI's Codex Can Now Train Its Own Open-Source Models, But Should It?

Hugging Face Blog · Dec 11, 2025 · 2 min read

OpenAI is handing its coding agent, Codex, the keys to the Hugging Face kingdom, and the implications are a bit dizzyin...

Swift devs can now share Hugging Face caches with Python, ending duplicate downloads — Inblix summary
Research

Swift devs can now share Hugging Face caches with Python, ending duplicate downloads

Hugging Face Blog · Dec 5, 2025 · 2 min read

Apple platform developers who’ve wrestled with loading large language models in their apps just got a major quality-of-...

Claude can now fine-tune open-source LLMs directly on cloud GPUs — Inblix summary
Research

Claude can now fine-tune open-source LLMs directly on cloud GPUs

Hugging Face Blog · Dec 4, 2025 · 2 min read

Anthropic's Claude just got a practical new ability that sidesteps the usual notebook grind: it can now fine-tune open-...

Intel's DeepMath slashes AI reasoning bloat by 66% using tiny Python snippets — Inblix summary
Research

Intel's DeepMath slashes AI reasoning bloat by 66% using tiny Python snippets

Hugging Face Blog · Dec 4, 2025 · 2 min read

Intel researchers have built DeepMath, a math agent that trades the long-winded, error-prone text chains of typical LLM...

Hugging Face Transformers v5 sheds TensorFlow, Flax, and 'slow' tokenizers in push for simplicity — Inblix summary
Research

Hugging Face Transformers v5 sheds TensorFlow, Flax, and 'slow' tokenizers in push for simplicity

Hugging Face Blog · Dec 1, 2025 · 2 min read

The Hugging Face team launched Transformers v5 today, marking a deliberate shift from the sprawl of 400+ model architec...

FLUX.2 demands 80GB VRAM but can now run on a 24GB GPU — here's how — Inblix summary
Research

FLUX.2 demands 80GB VRAM but can now run on a 24GB GPU — here's how

Hugging Face Blog · Nov 25, 2025 · 2 min read

Black Forest Labs just dropped FLUX.2, a new open image generation model that is decidedly not a simple upgrade to FLUX...

How continuous batching squeezes every drop of throughput from your AI chatbot — Inblix summary
Research

How continuous batching squeezes every drop of throughput from your AI chatbot

Hugging Face Blog · Nov 25, 2025 · 2 min read

If you’ve ever watched a chatbot like Qwen or Claude compose a response, you’ve seen the bottleneck: a long pause, then...

Tavily's Agent Rebuild Reveals a Brutal Truth: Your AI's Context Window Is a Ticking Time Bomb — Inblix summary
Research

Tavily's Agent Rebuild Reveals a Brutal Truth: Your AI's Context Window Is a Ticking Time Bomb

Hugging Face Blog · Nov 24, 2025 · 2 min read

Tavily’s team learned a painful lesson that many AI engineers are just beginning to face: building an agent harness tha...

OVHcloud brings sub-200ms inference to Hugging Face, starting at €0.04 per million tokens — Inblix summary
Research

OVHcloud brings sub-200ms inference to Hugging Face, starting at €0.04 per million tokens

Hugging Face Blog · Nov 24, 2025 · 2 min read

Hugging Face is expanding its Inference Providers ecosystem with OVHcloud, bringing a distinctly European option to the...

NVIDIA's Conformer-LLM hybrid tops ASR accuracy, but open-source still trails in long-form audio — Inblix summary
Research

NVIDIA's Conformer-LLM hybrid tops ASR accuracy, but open-source still trails in long-form audio

Hugging Face Blog · Nov 21, 2025 · 2 min read

The Open ASR Leaderboard just got a major expansion, adding multilingual and long-form transcription tracks that reveal...

RapidFire AI claims 20x faster LLM fine-tuning by running configs concurrently on a single GPU — Inblix summary
Research

RapidFire AI claims 20x faster LLM fine-tuning by running configs concurrently on a single GPU

Hugging Face Blog · Nov 21, 2025 · 2 min read

Fine-tuning large language models is a slog of trial and error. You tweak a learning rate, wait hours for a full run, s...

AnyLanguageModel swaps one import to unify Core ML, MLX, and cloud LLMs on Apple devices — Inblix summary
Research

AnyLanguageModel swaps one import to unify Core ML, MLX, and cloud LLMs on Apple devices

Hugging Face Blog · Nov 20, 2025 · 2 min read

iOS and macOS developers building AI features face a messy reality: local models use Core ML or MLX APIs, cloud provide...

The counterintuitive data trick that makes Mamba hybrids actually reason — Inblix summary
Research

The counterintuitive data trick that makes Mamba hybrids actually reason

Hugging Face Blog · Nov 19, 2025 · 2 min read

The team behind Apriel-H1 discovered something that should make AI engineers reconsider their default approach to disti...

Hugging Face Now Lets You Build and Share AMD ROCm Kernels Without the Toolchain Headaches — Inblix summary
Research

Hugging Face Now Lets You Build and Share AMD ROCm Kernels Without the Toolchain Headaches

Hugging Face Blog · Nov 17, 2025 · 2 min read

If you've ever wrestled with CMake, Nix, or ABI issues just to get a custom GPU kernel running in PyTorch, the new work...

AMD drops $10K prize for the best SO-101 robot hack in Tokyo and Paris — Inblix summary
Research

AMD drops $10K prize for the best SO-101 robot hack in Tokyo and Paris

Hugging Face Blog · Nov 13, 2025 · 2 min read

AMD is putting real money — and real silicon — behind the next wave of physical AI. The chipmaker, alongside Hugging Fa...

Hugging Face and Google Cloud build a CDN to slash open model download times — Inblix summary
Research

Hugging Face and Google Cloud build a CDN to slash open model download times

Hugging Face Blog · Nov 13, 2025 · 2 min read

The open model library Hugging Face and Google Cloud are deepening their tie-up with a concrete engineering fix for a m...