AI in week 45.

CM

Claude Mythos 5.2

Model · Anthropic

4

Anthropic's unrestricted variant of the Fable 5.2 model, available to vetted cybersecurity and biomedical organizations.

I2

Inkling 2 Small

Model · Thinking Machines

2

Open-weights multimodal MoE model (594B total, 17B active) from Thinking Machines, available via Tinker.

R6

Ripple 65B Preview

Model · Recursive Intelligence

2

Dense 65B open-weight language model from Ricursive Intelligence, trained for agentic coding with up to 1M context.

L2

LongCat 2.5

Model · Meituan

2

Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.

GC

GPT-6 Cyber

Model · OpenAI

2

OpenAI cybersecurity model for finding and validating exploitable vulnerabilities, with access limited to vetted defenders.

753B-parameter open-weights scientific-discovery model from DeepPrinciple, post-trained from GLM 5.3.

E5

ERNIE 5.5 Flash

Model · Baidu

1

Baidu's open-weights 280B-parameter sparse MoE base model (14B active) with 1M-token context, released pre-post-training.

Ant Group's multimodal mixture-of-experts model (560B total, 25B active) in the Ling family, served via API.

Q3

Qwen 3.5 0.8B

Model · Alibaba

1

Small 0.8B-parameter open-weights language model in Alibaba's Qwen 3.5 family, aimed at lightweight and on-device use.

GI

GPT Image 2.6

Model · OpenAI

3

OpenAI image generation and editing model with GIF output, SVG export and image markup, in Flare and Sunburst variants.

PI

P-H3 Image

Model · Pruna AI

3

Open-weight image generation and editing model from Pruna AI, distilled from MiniMax's H3 video model to run in 8 steps.

PJ

PJiT

Model · Linum AI

1

2B-parameter pixel-space text-to-image diffusion model from Linum AI, trained from scratch for about $21K; Apache 2.0.

N2

NitroFusion 2

Model

1

One- and two-step image generation model distilled from Z-Image-Turbo using text-stream adversarial distillation.

GO

Gemini Omni 1.2 Flash

Model · Google

3

Google's video generation model producing clips up to 15 seconds from text, image, video or audio input.

PH

Pika H3 Flex

Model · Pika Labs

2

Pika's open-weight video model post-trained on MiniMax H3, generating synchronized video and audio up to 4K.

DR

DreamX

Model

1

Research model that turns a pretrained video diffusion model into a latent action learner for human-to-robot motion transfer.

RE

RelayVSR

Model

1

Streaming real-world video super-resolution using sparse large-model keyframes and a lightweight propagation network.

Early-access AI video editing app from Steve Aoki and a partner that learns a creator's editing taste.

MO

MOSS-Live

Model

1

Full-duplex real-time voice, video and screen assistant from MOSS (模思智能) with background task execution.

AV

Apriel Voice 27B

Model · ServiceNow

1

Open-weights 27B speech-to-text model from ServiceNow that adds a speech interface to a Gemma 4 backbone, with tool calling.

CO

CoLa3D

Model

1

Research model that decomposes a single image or scene mesh into object-level 3D meshes with consistent spatial layout.

3D reconstruction model that recovers geometry and camera poses from photos or video using a loop transformer.

MO

Mobius

Model

1

Unified motion-language model for text-to-motion, motion captioning, completion and editing, from HKUST and Tencent Games.

HA

HairDeck

Model

1

Research model that generates animation-ready stylized hair cards and spring-bone chains from a single character image.

FL

FlowHMR

Model

1

Monocular video motion capture that combines flow matching with physics-based RL for plausible 3D human motion.

RE

ReLift

Model · Nvidia

1

Feed-forward model that repairs lifted 2D tracks and depth into accurate dense 3D trajectories from monocular video.

LL

Lucid Lyra

Model · Nvidia

1

NVIDIA research model that turns a single image into a queryable 3D Gaussian scene with semantics and object instances.

VO

VoxLution

Model

1

Research model generating 3D indoor scenes as sparse voxels, refined from 25 cm to 1.5625 cm by an autoregressive cascade.

NO

NorthCUA

Model · Beipo AI

1

Open-weights computer-use model family (9B, 35B, 397B MoE) from Beipo AI, built on Qwen3.5, with an open-source harness.

CA

ChatGPT and Codex shared usage limits

App · OpenAI

1

OpenAI policy under which ChatGPT and Codex usage allowances are drawn from a single shared pool on a ChatGPT plan.

QW

Qwen3.5-4B-GLB-3D-Coder

Model · Community (ccore)

1

Apache 2.0 fine-tune of Qwen3.5-4B that writes Three.js pages for low-poly 3D assets with GLB export.

CB

Claude bank account connection

App · Anthropic

1

Read-only bank account connection in Claude via Plaid, for answering questions about spending and finances.

Creator-built custom tool for tracking YouTube video stats and outlier videos, shown by Matthew Berman.

EC

Excel clone

App

1

Browser-based spreadsheet app ("Quadrille Spreadsheet") built with AI models as an Excel clone.

FO

Folio

App

1

Lightweight PDF reader and editor with annotation, e-signing and AI Q&A, positioned as an Adobe Acrobat alternative.

ZA

ZAP

Model

1

27B vision-language robot policy trained only in simulation, run zero-shot on a real bimanual robot.

Open-source robotics toolkit of four libraries: motion planning, motion retargeting, grasping and agent skills.

MIT research showing a frontier LLM can read and write another neural network's raw weights without gradient descent.

Tiny ~51M-parameter vision model from Cactus Compute for on-device object detection, text reading and spatial understanding.

ST

STRATA

Model · Nvidia

1

Nvidia's 332M-parameter AI model emulating the SCREAM global storm-resolving atmosphere simulation at ~4.9 km, 10-minute steps.

MI

METR Independent Assessment of AI R&D Acceleration at Anthropic

Research · METR

1

METR report assessing how much AI is accelerating AI R&D inside Anthropic, based on high-access review (October 2026).

PE

Perplexity Embed V2 Context

Model · Perplexity

1

Open-weights vision-language embedding models from Perplexity (9B and 600M) for text, image and document retrieval.

FF

FFmpeg

Library

1

Open-source command-line toolkit and libraries to record, convert, filter and stream audio and video.

YT

yt-dlp

Library

1

Command-line audio/video downloader supporting thousands of sites, forked from youtube-dl.

NA

Netflix AI Discover tab

App · Netflix

1

AI-driven Discover tab in the Netflix app that groups recommendations by mood and viewing occasion.

OM

Omarchy

App

1

Open-source, opinionated Linux distribution by DHH with built-in AI agent integration for debugging and customization.