Open Weights
Qihoo 360's LLM family
Open-weights LLM lab founded by Sogou's Wang Xiaochuan
Open Japanese LLMs from Japan's biggest ad group — Apache 2.0
Japan's Japanese-first LLM — from U Tokyo's Matsuo Lab, backed by KDDI
Makers of Mochi 1, the open-weights text-to-video model
Free chat across 122 open models, auto-routed by Omni
Tencent's image-generation model — open-weights
Tencent's open-weights text-to-video model
UAE's open-weight Arabic LLM family, now Jais 2 at 70B params
Kakao's on-device Korean AI — hyper-personalised chat assistant built into KakaoTalk for 54 million users
India's first AI unicorn — multilingual LLMs and GPU cloud built for 1.4 billion people
Meta's open-weight LLM family — Llama 4 landed, Behemoth didn't
Meta's open classifier for filtering unsafe LLM prompts/replies
European frontier-model lab
NVIDIA's open-weight LLM family tuned for agentic workloads
Open, efficient small language models from Microsoft Research.
Alibaba's LLM family: open Qwen3 weights plus closed Qwen3.7-Max
Open Arabic LLMs and dialect-aware TTS, free-weighted on Hugging Face
Abacus.AI's 2024 open-weight fine-tunes, now legacy
High-performance inference for open-weights LLMs
Korea's first gaming-industry LLM — bilingual Korean-English AI for creative, enterprise, and multimodal applications.
Mistral's open-weight TTS that beat ElevenLabs in blind tests
01.AI's Yi model family — frontier development paused, company pivoted to enterprise
Open-weights frontier LLM family from Zhipu AI
Saudi Arabia's humain-m3 Is MiniMax's M3, Under MiniMax's Licence
Same 428 billion parameters, same 23 billion active, same mixture-of-experts design as the model Shanghai open...
KE
5 Sep
Japan's Banks Waited Eleven Weeks for a Capability Anyone Can Download
A US directive covered foreign nationals, so Anthropic switched the models off for everyone, including America...
MA
31 Aug
Meta Promised Frontier Weights Three Weeks Ago. The Repository Is Still Empty.
Zuckerberg promised the weights for Meta's frontier model three weeks ago, with no date attached. We queried t...
MA
31 Aug
The Anonymous Model Called Ox Alpha Was Z.ai's, and Its Weights Are MIT
An unattributed model ran free on OpenRouter and OpenCode for six days before researchers fingerprinted it and...
MA
29 Aug
Moonshot Wants 30% of What the Clouds Earn From Its Model
The usual arrangement runs the other way: the platform takes a cut from the model. Asking Microsoft, Amazon an...
MA
27 Aug
Nvidia Paid $7bn For Poolside's Technology And Staff Without Buying Poolside
A US$6bn non-exclusive licence to the model factory, 109 employees transferring, and US$1bn of equity at a US$...
EV
22 Aug
AMD Is Buying a Chip That Runs One Model. The Number That Decides It Is Two Months.
AMD is buying Taalas, a Toronto startup that etches a trained model's weights directly into silicon rather tha...
AI
9 Aug
Alibaba Published Qwen3.8-Max's Benchmarks. Every One of Them Is Alibaba's.
Alibaba made Qwen3.8-Max generally available on 3 August and published benchmark scores for the first time. Ev...
AI
8 Aug
Kimi K3's Weights Are Free — the 1,561 GB You Need to Hold Them Are Not
Moonshot published 2.8 trillion parameters on 27 July. We summed all 96 shards: the file sizes independently c...
AI
28 Jul