FOUNDATION MODELS
Foundation Models
Open source · Agents · Multimodal
40 articles
- Shanghai AI Lab's 8B model quietly rewrote the math on training cost — 2026-10-28
- OpenAI's open weights just gave Chinese developers another local option — 2026-10-28
- Baidu open-sources an 8B image model that runs on a consumer GPU — 2026-10-28
- A Chinese speech model just out-ranked Whisper on Hugging Face — 2026-10-28
- Qwen3 ships hybrid thinking and eight open-weight models at once — 2026-10-27
- Volcano Engine splits model access into four tiers with steep off-peak discounts — 2026-10-27
- Ant Group's Ming-Omni folds text, image, audio and video into one open model — 2026-10-27
- Kimi K3.1 Pro puts a 3-trillion-parameter open-weight model within three points of the frontier — 2026-10-27
- A Chinese research model helped prove a 30-year-old math conjecture — 2026-10-27
- Kimi K2: Moonshot open-sources a 1-trillion-parameter agent model — 2026-10-27
- GLM-4.5 unifies reasoning, coding, and agent skills in one open model — 2026-10-27
- DeepSeek-R1 proved open weights could reason like o1 — 2026-10-27
- Xiaomi ships MiMo-V2.6 as a native omnimodal model and open-sources it — 2026-10-25
- SenseTime's SenseNova V6 banks on native multimodal to top two leaderboards — 2026-10-25
- Meituan open-sources LongCat-2.0, a trillion-parameter model built on 50,000 domestic chips — 2026-10-25
- Ant open-sources Ling-3.0-flash-Fin, a 124B model that runs finance workflows on 5.1B active params — 2026-10-25
- Qwen-Image makes multilingual text rendering an open-weight strength — 2026-10-24
- Can a 10B-active model punch like a 230B? MiniMax M2 makes the case — 2026-10-24
- Tencent's Hunyuan3D 2.0 turns a single image into a textured 3D model, openly — 2026-10-24
- DeepSeek V3.2 bets on sparse attention for cheaper long context — 2026-10-24
- Step 5 Preview goes open-weight and lands at number two on OpenRouter — 2026-10-23
- iFlytek bets on fully domestic compute with Spark X2.6 and an end-cloud split — 2026-10-23
- Tencent rebuilds Hunyuan from the ground up and open-sources Hy3 preview — 2026-10-23
- Doubao-Seed-2.1 Pro turns coding and agents into a volume game — 2026-10-23
- Alibaba's new model listens, watches, and starts working — 2026-10-22
- Huawei open-sources a model built for its own chips — 2026-10-22
- Manus returns with an agent that finishes the job — 2026-10-22
- Baidu's new model does more on a fraction of the training bill — 2026-10-22
- Step 3: a 321B open multimodal model designed to decode cheaply on Chinese chips — 2026-10-21
- MiniMax-M1: a 456B reasoning model with a 1-million-token window trained for $534,700 — 2026-10-21
- Kimi K2: a 1-trillion-parameter model that put a Chinese lab on top of the open-source leaderboard — 2026-10-21
- GLM-4.5: Zhipu's agent-native 355B model that ships under a permissive MIT license — 2026-10-21
- Can a 600-billion-parameter model keep pace with the trillion-parameter giants? — 2026-10-20
- MiniMax M2 bets that a compact coding model beats a giant on cost — 2026-10-20
- DeepSeek V4 Preview makes a million-token context the default, not the premium — 2026-10-20
- ByteDance's Seed 2.1 targets production coding, not demo chat — 2026-10-20
- Zhipu's GLM-4.6 makes open-weight coding agents run on domestic chips — 2026-10-19
- Kuaishou's Kling 4.0 takes AI video to 30 seconds and 10-bit HDR — 2026-10-19
- Moonshot's Kimi K2.5 deploys 100 sub-agents on one task — the agent-swarm bet — 2026-10-19
- Tencent's Hunyuan Hy3 topped OpenRouter in a week — a quiet bet on useful AI — 2026-10-19