AI in week 44.

Open-weights diffusion language model from DiffuSpace AI with 35B total and 3B active parameters.

LF

LFM3

Model · Liquid AI

2

Liquid AI's LFM3 family of small multimodal foundation models for on-device, edge and cloud use.

33B dense open-weights LLM from Nanbeige, tuned for agentic coding and professional cowork tasks.

A3

Apex 35B

Model · Apexin

1

35B-parameter model from Apexin, built on Qwen3.6-35B-A3B and tuned for automated research and robotics simulation.

Family of four open MoE language models (19B to 280B), reportedly pruned from DeepSeek and Qwen bases.

SP

Spark-X2.6-1.7B

Model

1

Compact 1.7B open-weights text LLM for on-device use, with adjustable thinking levels and English/Chinese support.

S9

SenseNova-Loop 9B

Model · SenseTime

1

9B-parameter looped multimodal model from SenseTime that reuses a layer block twice for long-context and agentic tasks.

GP

GPT-6.1

Model · OpenAI

1

OpenAI language model in the GPT-6.1 family; the Soul Max variant is noted by reviewers for low token use per task.

Proof-of-concept image generator that learns an energy function and lowers it iteratively instead of denoising.

MA

MatAnyone-E

Model

1

Efficient high-resolution video matting model using sparse memory propagation, extending MatAnyone 2.

UN

UniAction

Model

1

Research framework that conditions one video world model on force, camera, hand pose and robot pose actions.

OP

Open-H3-Regenerate-2K

Model · NVIDIA

1

Open few-step diffusion refiner from NVIDIA that regenerates low-resolution H3 video drafts at 768P to 4K, 5–15 seconds.

S6

SenseNova 6.8 Code2Video

Model · SenseTime

1

Open-weights 27B model from SenseTime that creates short videos by generating and rendering executable code.

EL

ElevenLabs

API · ElevenLabs

2

AI audio platform and API for text-to-speech, speech-to-text, music, sound effects, voice cloning and voice agents.

HY

Hy-SimulMT

Model · Tencent

1

Tencent Hunyuan's real-time Chinese-English speech-to-speech simultaneous translation model with voice cloning.

AT

Audio8 TTS Realistic 0.6B

Model

1

Compact 0.6B open text-to-speech model with zero-shot voice cloning and support for 23 languages.

P1

Paradee 1.2

Model

1

8.5 MB, 8M-parameter text-to-speech model distilled from Kokoro-82M, with four preset voices and CPU inference.

NN

NVIDIA Nemotron 3.5 Transcribe

Model · NVIDIA

1

NVIDIA's open-weight mixture-of-experts speech model for offline transcription in 40 languages, up to one hour of audio.

G4

Genie 4

Model · Google DeepMind

3

Google DeepMind world model that turns text or images into interactive, real-time controllable worlds with audio.

MO

MotionBricks

Model · NVIDIA

2

NVIDIA's real-time generative motion model for game characters and humanoid robots, with smart primitives for authoring.

MU

MUNDO

Model

1

World model that turns a few context images into an editable, physics-simulated 3D world rendered by a video model.

OV

OverSeer

Research · Meta

1

Meta/Oxford multi-agent video world model that adds fixed auxiliary cameras to keep a shared world consistent.

PH

PhysAg3D

Research

1

Research framework that turns a single image into a textured 3D asset with articulated parts and physical behavior.

ID

IDOL

Model

1

Diffusion model from Speridlabs that turns a single image into a 3D Gaussian splat object at adjustable detail levels.

SC

SceneBloom

Research

1

Research framework that generates collision-free, physically stable 3D indoor rooms from diffusion-sampled layouts.

AU

AutoSwarm

Agent · HKUDS

1

Open-source layer that coordinates multiple coding and general agents (Codex, Claude Code, etc.) in parallel with shared context.

ST

StreamVibe

Agent

1

Research project on proactive streaming video agents that infer implicit, evolving user intent and decide when to help.

RR

Roast Room

App · Nate Herk

1

Open-source Claude Code skill and local web app where AI advisor personas debate a business decision and output a PDF verdict.

Claude Code skill that turns an X post link into a researched, fact-checked Instagram carousel and posts it automatically.

HE

Herk Edit Kit

Library · Nate Herk

1

Open-source kit of 15 video-editing skills for Claude Code and Codex, built on HyperFrames and GSAP.

BL

Blotato

App · Blotato

1

Social media publishing tool, API and MCP server for scheduling, DMs and analytics across 9 platforms, with AI content generation.

NVIDIA dataset of 2,600+ simulation-ready 3D objects with physical properties for robotics and simulation training.

OS

OpenAI scientific discoveries

Research · OpenAI

2

OpenAI's announced AI-produced results: a local smoothing proof, an ETH refutation and a QCD result, with Lean verification.

AA

Artificial Analysis Intelligence Index v5

Research · Artificial Analysis

2

Artificial Analysis's aggregate benchmark index that scores language models on intelligence across multiple evaluations.

Google research training framework alternating wake and sleep phases for memory consolidation and self-generated practice.

UO

Unlimited OCR 2

Model · Baidu

2

Baidu's open-source OCR model that parses very long multi-page documents in one pass with near-constant decoding cost.