FlappyBench Showdown: DeepSeek V4.1-Flash Wins on Cost, GLM 5.3 on UI, Kimi K3 Lags
A same-prompt FlappyBench run shows DeepSeek V4.1-Flash delivers 9/10 at the lowest one-shot cost, GLM 5.3 scores 9.5/10 with a polished UI at 2x cost, and Kimi…
A same-prompt FlappyBench run shows DeepSeek V4.1-Flash delivers 9/10 at the lowest one-shot cost, GLM 5.3 scores 9.5/10 with a polished UI at 2x cost, and Kimi…
Meituan LongCat welcomed Hugging Face to its Shanghai office for talks on future model development, open-source plans, and the latest Transformers features.
Requesty's production data shows teams explore more models over time while weekly model usage stays stable, suggesting model choice is an ongoing process rather…
Hugging Face has finished its six-session Training Agents series, with all recordings on YouTube and code open-sourced. The series covers agent evaluation to tr…
Google DeepMind used restored archival photos and pose control models to recreate the first meeting of Burt and Ethelle for the documentary 'Love, Rendered', ma…
Hugging Face announces an agent-swarm challenge for the Hutter Prize, inviting AI agents to collaborate on finding better compression algorithms via Codex, Clau…
Ling 3.0 Flash VL is now available on OpenRouter, extending InclusionAI's Ling 3.0 Flash with stronger language performance, native image/video input, and advan…
Ling 3.0 Flash VL, a 124B MoE with 5.5B active parameters, is now available on OpenRouter via Novita Labs, adding native image/video input and visual agent capa…
Qoder highlights a user-built short film demonstrating a new cross-device workflow: desktop Qoder generates a Suqian travel guide PDF that syncs to mobile Qoder…
A developer explains why they switched from Opencode to Command Code, praising the project's steady updates after its open-source model launch and rallying the…