AI Model Pulse

Fenghe, a domain-specific foundation model from the China Meteorological Administration, is trained on 50 million meteorological tokens to support...

• 2.4T-parameter multimodal model debuts as Alibaba's largest yet, positioned second only to Anthropic's Fable 5.
• No benchmarks or model cards...

Moonshot AI's Kimi K3 launch on July 17 marks a major shift: the 2.8-trillion-parameter open-weight model matches leading OpenAI and Anthropic systems...

DiffusionGemma, a new 26B text-diffusion model from Google DeepMind, can fix its own wrong answers mid-generation. Announced at RAAIS 2026 just 48 hours after shipping, it signals a potential departure from autoregressive limits.

New Model Releases

• 🔥 Kimi K3: Moonshot AI released Kimi K3, a 2.8 trillion parameter MoE model with 1M token context that activates 16 of 896...

Alibaba announced Qwen 3.8 (2.4T params, multimodal) as 'open-weight soon' via X, with a preview live on Token Plan and Qoder but no benchmarks,...

Why launch a 975B-parameter model that explicitly refuses to claim top benchmarks?

• Architecture enables the bet: DeepSeek-V3-inspired MoE activates...

Moonshot's Kimi K3 upends assumptions of a widening US-China gap with a 2.8T-parameter model that trails only Claude Fable 5 and GPT-5.6 Sol.

-...

New Open-Weight Releases

• 🔥 Kimi K3: Moonshot AI released Kimi K3, an open-source model demonstrating frontier-level performance across...

Chinese labs dominated recent open-weight releases at the 2026 World AI Conference.

• Kimi K3 (Moonshot AI) launched with 2.8 trillion parameters,...

New Open-Weight Releases

• 🔥 Inkling: Thinking Machines Lab released Inkling, a 975B-parameter mixture-of-experts model with 41B active...

VideoChat3 introduces a fully open 4B-parameter video MLLM that balances broad generalization with high efficiency via I3D-ViT and adaptive frame...

Google postponed Gemini 3.5 Pro by several months after its code-generation performance missed internal targets.

• Alphabet shares dropped 4% on the...

Thinking Machines Lab released Inkling, a 975B-parameter MoE model with 41B active parameters trained on 45 trillion multimodal tokens, but explicitly...

• Technical scale: Moonshot released the world's largest open-weight model at 2.8T parameters with a 1M-token window, delivering competitive...

Frontier Lab Releases

• 🔥 Inkling: Thinking Machines Lab released Inkling, a 975B-parameter open-weight multimodal MoE model with 41B active...

• Thinking Machines Lab released Inkling as its first open-weights model: a 975B-parameter MoE with 41B active params, native text/image/audio...

• Coding prowess: Achieves 62.1% on SWE-Bench Pro and 74.4% on FrontierSWE, close behind Claude Opus 4.8 while beating GPT-5.5 on select benchmarks
  -...

Nvidia introduced Cosmos 3 Edge, a world model enabling robots and vision AI agents to perceive and navigate physical environments in real time,...

No significant updates today.