Tag
Qwen is teasing an upcoming release of the Qwen3.8-27B model on Hugging Face, with a countdown to August 14, 2026, promising 'unmatched intelligence density'.
Qwen3.8-27B, a compact 27B dense vision-language model with flexible thinking control and long-context support, is released as the most capable Qwen open model to date, available soon via Hugging Face and Qwen Cloud.
Rohan Paul's daily AI newsletter digest covering SpaceXAI's Grok 4.6 launch, Soniox TTS v2, NVIDIA Nemotron 3.5 Lightning, OpenAI's ChatGPT Linux app, and several research papers.
Announces the release on Huggingface of Qwen3.8-2.4T-A95B, a 2.4T-parameter open-weight mixture-of-experts model.
Grok 4.6 is officially released, with the same pricing as 4.5, enhanced agent and programming capabilities, and it runs longer on complex tasks. The AA Intelligence Index ties GPT-5.6 Sol, with input at $2 per million tokens and output at $6 per million tokens.
Simon Willison covers the release of DeepSeek V4 Pro 0813, available via API on OpenRouter, noting the likely open weights and noticeable differences between reasoning levels.
用户感叹开通 Premium+ 赠送的 SuperGrok 价值上升，引用 Elon Musk 宣布 Grok 4.6 发布，称其智能、快速且性价比高。
Grok's latest model, grok-4.6-high, ranks #43 on LMArena text and 7th on webdev, failing to reclaim the top spot Grok last held in December 2025.
Qwen3.8 Max, the largest Qwen model, is now available on Hugging Face with 2.4T parameters, 95B active MoE, 1M context, and FP8 version, open for developers and companies.
Announcement of today's releases of Grok 4.6, DeepSeek v4 Pro, and open-source Qwen 3.8 models.
DeepSeek V4 Pro 0813 is now available with API pricing details on OpenRouter, marking a new model release for developers.
Deepseev v4pro 0813 is now rolling out to the API, marking a new model update for developers.
DeepSeek announces that DeepSeek-V4-Pro-0813 is now available.
Qwen 3.8 2.4T has been released on Hugging Face, while the 27B variant is not available yet and is expected later.
This post clarifies the exact release date and time for Qwen 3.8 27B, linking to the official model page on ModelScope.
A newly released large MoE model, Motif 3 (314B with 13B active parameters, NVFP4 available), seems promising; the author asks for community experiences.
Announces the imminent release of Qwen in just over 7 hours, calling it a historic event and hinting that Google Translate is not ready for it.
Microsoft's MAI-Code-1.1-Flash coding model is rolling out in GitHub Copilot, adding native vision support and improved coding performance at a 73% lower list price than its predecessor.
NVIDIA announced Nemotron 3.5 Lightning, a 30B mixture-of-experts open model optimized for high-volume agentic AI workloads, alongside NeMo Switchyard, an open-source library for intelligent model routing across heterogeneous model ecosystems.
Aaron Levie discusses how open-weight frontier-class models will unlock AI adoption in regulated and private-infrastructure scenarios, citing Alexandr Wang's announcement of Muse Spark 1.2 and Muse Glimmer, a 30B agentic model under Apache 2.0.