Tag
作者做了一个实验,让指令微调的 LLM 反向生成"用户轮次"(user_start token 之后的文本),测试约 30 个模型后发现用 chat template 反转技巧比角色扮演 system prompt 更能模仿真人,其中 Mistral Small 3 24B 表现最佳。结果和代码已开源在 selfgpt 网站与 GitHub 仓库。
The author praises Mistral's revamped chat app 'Vibe Chat' (formerly Le Chat), highlighting its default Work/Cowork-like mode with native file creation and analysis, strong scheduled job integrations, a generous free tier, and superior conversational quality for everyday queries compared to ChatGPT, Claude, and Gemini.
Mistral 发布 Voxtral 3 语音模型家族与语音代理 SDK,支持 50+ 语言的实时转录、情感自适应语音合成及多步工具调用,可通过欧盟托管 API 或自部署使用以保障数据主权。
CERN and Mistral have signed a statement of intent for a long-term partnership to build a Large Physics Model — the first foundation model for collider physics — trained on billions of simulated LHC proton-proton collisions to support tasks from particle reconstruction to anomaly detection.
Mistral 发布 Timestral 时间序列基础模型家族,包含 5M 到 209M 参数四个版本,其 209M 模型在 GIFT-Eval、TIME 和 fev-bench 三大基准上均排名第一,支持零样本概率多变量预测且可在 CPU 上推理。
Analysis of the birthplaces of the top 30 most-cited AI researchers shows strong global talent distribution, yet no European company ranks among the top 40 largest public AI firms by market cap, highlighting the EU's struggle to convert research talent into commercial AI leadership despite Mistral's presence.
Mistral 发布 Timestral 时间序列基础模型家族(5M/28M/123M/209M 参数),支持零样本预测、单变量与多变量预测及概率输出,209M 版本代码与权重已在 GitHub 和 Hugging Face 开源。
Mistral AI 发布 Timestral 时间序列基础模型系列(5M/28M/123M/209M),209M 版本在 GIFT-Eval、TIME 和 fev-bench 上实现零样本预测的最先进精度,并开源代码。
A roundup of four notable AI news items: OpenAI's mathematical research output surpassing 700 papers, Mistral's release of the Le Chonk open-source model, GPT-6 Astra's website-comparison capability, and Hark's launch of an active AI assistant by Brett Adcock.
Mistral 发布了拥有 1 万亿参数的新模型 Mistral Large 4(绰号 Le Chonk),声称可与美中顶尖 AI 模型竞争,是目前中国以外最强的开源权重模型,特别针对编码、网络防御及制造业、金融等垂直领域优化,现处于预览阶段。
Mistral has released its flagship Mistral Large 4, with open weights promised at the end of the month, reasserting Europe's position in the frontier AI race. The release is accompanied by a broad research portfolio spanning reasoning, multimodality, audio, coding, safety, and physical AI.
The llm-mistral plugin for Simon Willison's llm CLI tool has been updated to version 0.16, adding support for Mistral models via the plugin interface.
Mistral has released a preview of Mistral Large 4, a 1-trillion-parameter (49B active) model trained on 3,800 NVIDIA Grace Blackwell GPUs, with open weights promised later this month; it scores 38 on Artificial Analysis, a huge leap over Mistral Large 3.
Simon Willison tests Mistral Large 4 alongside frontier models (Claude Opus 5.5, GPT-6.1, Gemini 3.8 Flash) by generating an SVG of an armadillo in fishnet tights jaywalking on Mars, highlighting how saturated benchmarks push toward absurd creative prompts.
Mistral Large 4 reportedly tops a benchmark related to EU AI regulation compliance, highlighting how regional policy considerations are becoming a competitive factor for European AI models like Mistral's.
Mistral AI has released Mistral Large 4 (ML4), nicknamed "Le Chonk", a 1-trillion-parameter multimodal model trained on just 4,000 NVIDIA GPUs, aiming to compete with both closed US models and Chinese open-weight rivals, with open weights planned in three weeks after safety testing.
Clement Delangue teases the upcoming Mistral Large 4 ("Le Chonk"), a 1-trillion-parameter natively multimodal model with 49B active parameters per token (52B including embeddings/output layers), with open weights expected on Hugging Face at the end of the month.
Mistral AI team member announces that Mistral Large 4 is still in preview, with continued reinforcement learning runs yielding further improvements, and the full release is expected by the end of the month.
Mistral AI announced Mistral Large 4, a state-of-the-art open-weight, general-purpose multimodal model with a granular MoE architecture featuring 49B active parameters, 1.05T total parameters, a 1.6B vision encoder, and a 1M token context window.
Mistral AI appears to have released benchmarks for its new Mistral Large 4 model. Details are minimal, but the post signals a significant model release worth watching for its performance numbers.