Kimi K3.1 Pro: 3T-Class Open-Weight Model With 1M-Token Context
Moonshot AI released Kimi K3.1 Pro, a 3T-class open-weight MoE model with 1M-token context, multimodal input, full-project coding, 3D modeling, and a 65% reduct…
Global AI Digest
-
-
Gemini Magic Pointer Arrives on the New Googlebook
Android announced Magic Pointer on the new Googlebook, which integrates Gemini directly into the pointer so users can wiggle and click across screens to ask que…
-
NVIDIA Nemotron 3.5 Bolt Vision-Language Model Now on Ollama
NVIDIA's Nemotron 3.5 Bolt, a small vision-language model for fast local conversations, image understanding, and tool use, is now available on Ollama with a 128…
-
Tencent Hy Releases Hy-Nomas, an Open-Source Interactive Video World Model
Tencent Hy launched Hy-Nomas, an open-source free-form interactive video world model supporting text prompts and camera trajectories, long-horizon generation wi…
-
MiniMax M3.1 Flash Open Weights Released, Available on GMI Cloud and OpenRouter
MiniMax released the open weights of MiniMax M3.1 Flash, a 428B-parameter model with 23B activated per token, and GMI Cloud made it available from day one via i…
-
OpenRouter Data: GLM 5.3 and Kimi K3 Lead Security Task Spend, Grok 4.7 Top Closed-Weight Model
OpenRouter shares a spend-based breakdown of models used for security tasks, with GLM 5.3 and Kimi K3 leading overall and Grok 4.7 the top closed-weight model.
-
Cursor Adds Computer Use So Agents Can Control Desktop Apps
Cursor introduced Computer Use, letting its agents click and type across any app on the user's computer, not just the browser, and drive desktop apps like the i…
-
Scenario Open-Sources 1,300+ 3D Models on Hugging Face
Scenario released over 1,300 3D models (39 GB) on Hugging Face, generated by tools including Hitem3D, Meshy, Rodin, Tripo, Claude Opus 5.5 3D and GPT-6 Astra 3D…
-
Requesty Reports 4ms Median Guardrail Latency
Requesty published latency figures for its AI gateway guardrails from Oct 3–18: 4 ms median, 15 ms p95, and 23 ms p95 on prompts over 64k tokens.
-
DeepSeek V4.1 Flash 1017 vs GPT-6 Astra: Same Three.js Prompt, 50% Lower Cost
SiliconFlow ran the same Three.js 3D-world prompt through DeepSeek V4.1 Flash 1017 and GPT-6 Astra, reporting nearly equal or better output quality at over 50%…
