Luma just released Uni-1, an image generation model that reasons first! The shift: image generation models typically wor...
Doubao-Seed-2.0-lite 0428 内测版本升级,新增音频理解功能,能同时支持图片、视频、音频和文本四种输入,成为豆包大模型家族首款全模态理解模型。该版本在 Agent、Coding 和 GUI 能力上也有明显提升。通过 API 测试,验证了其在前端动效复刻、视频Hooks建议和字幕识别等场景的应用潜力,具体案例详见后续推文线程。
BREAKING: MiMo-V2.5 by @Xiaomi is #6 overall out of open weight models on Design Arena! This has an Elo of 1297 and is i...
OpenAI 已全量发布 GPT-5.5 Instant 模型,并将其设为 ChatGPT 的默认模型。此次升级显著提升了模型的实时准确性和日常任务处理能力,特别是在法律、金融和医学等领域有效降低了幻觉率。同时,模型在图片理解与文档解析方面表现更佳。其回答风格变得更加简洁、聚焦要点,并增强了清晰度、个性化以及温暖自然的语调。此外,GPT 引入了记忆来源功能,允许用户可视化查看并编辑记忆的来源。本次更新未包含 Codex 模型。
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
Google发布了Gemma 4系列模型的专用草稿模型,用于推测性解码优化。31B Dense模型搭配草稿模型速度提升3倍,仅增加1G显存开销;Gemma4-26B和Gemma4-E4B分别提升1.5倍和3.1倍速度。新草稿模型如gemma-4-31B-it-assistant体积仅939 MB,专门优化后接受率高,相比之前使用非专用草稿模型(如gemma-4-E2B-it-UD-Q4_K_XL)提速更明显。作者呼吁Qwen尽快推出类似优化模型(如Qwen3.6-27B-assistant),以应对高性能需求。
前沿模型SubQ基于创新的Subquadratic Sparse Attention架构,实现了1200万token的实用上下文窗口。其核心技术SSA通过内容依赖的选择机制,让每个查询仅动态计算与相关键的注意力,使计算和内存成本随序列长度线性增长,而非传统Transformer的二次方增长。实测在100万token时比FlashAttention-2快52.2倍,成本低于Opus的5%。该模型针对需要一次性处理完整代码库、长文档等企业真实长上下文场景优化,旨在弥合“名义上下文”与“功能上下文”窗口的差距。
Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse...
Google通过为Gemma 4引入MTP drafters(多token预测草稿机),在不增加参数、不改变架构和模型质量的前提下,实现了最高3倍的推理速度提升。该技术让模型能一次预测多个token,突破了传统自回归解码的串行瓶颈,极大提升了GPU利用率。这显著增强了本地部署的实时性,并使Agent、代码生成等场景受益,进一步放大了开源模型在性价比和本地运行方面的优势。
Gemma 4: Now up to 3x Faster. ⚡ Same quality, way more speed. Our new MTP drafters allow Gemma 4 to predict multiple tok...
Inworld AI 发布了新一代实时对话语音模型 Realtime TTS-2,突破了传统TTS仅追求拟人化的竞争框架。该模型能够实时倾听完整对话,捕捉情绪、语气与节奏,动态决定回应方式,成为一个“会倾听、会共情、会适配”的对话伙伴。其关键创新包括:支持用自然语言语音指令像指挥大语言模型一样调整声音;同一声音身份可跨100多种语言保持一致性;还能通过文字描述生成全新声音并保存复用。这标志着语音AI从机械的“语音输出”迈向更贴近真人互动的“实时对话”时代。
Introducing Realtime TTS-2, a new generation of voice model built for realtime conversation. It is the first voice model...
OpenAI已将GPT-5.5 Instant设置为ChatGPT的默认模型。该模型在医学、法律和金融等关键提示上的虚假陈述比前代减少52.5%,在用户标记过事实错误的困难对话中不准确陈述降低37.3%。回答更简洁,用词和行数减少约30%。同时,模型增强了图像和STEM处理能力,并能从记忆、过往对话、文件及连接的Gmail中提取上下文以实现深度个性化。OpenAI表示,此次升级旨在提供更智能、清晰、个性化且语气自然温暖的回答。
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
Holy shit GPT 5.5 instant 1. Doesn't yap 2. Talks like a human 3. Brings relevant things into the convo (I was having ai...
Introducing Realtime TTS-2, a new generation of voice model built for realtime conversation. It is the first voice model...
the new instant model in chatgpt is so good damn if you have been thinking-model-only for awhile, give it a try!
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
Grok 4.3 is now live on the xAI API. It's our fastest, most intelligent model to date. It tops the @ArtificialAnlys lead...
Gemma 4: Now up to 3x Faster. ⚡ Same quality, way more speed. Our new MTP drafters allow Gemma 4 to predict multiple tok...
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
OpenAI 将 ChatGPT 默认模型全面升级为 GPT-5.5 Instant,替换原有的 GPT-5.3 Instant,并向所有用户推送。新模型显著减少了事实幻觉,在医疗、法律等高风险领域的错误率大幅降低,并在多项基准测试中性能提升。回答风格更为简洁,减少冗余内容。同时增强了记忆功能,能主动利用用户历史聊天记录等数据提供个性化回答,并上线“记忆来源”功能供用户管理引用。免费用户即可使用新版模型,付费用户可保留旧版三个月。个性化记忆功能将逐步向各版本用户开放。
GPT-5.5 Instant is starting to roll out to everyone in ChatGPT. Much more concise. Better memory. More personalized. And...
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
OpenAI正式将GPT-5.5 Instant设置为ChatGPT的新默认模型,该模型在基准测试中表现大幅提升,变得更智能、准确和可靠。其在图像分析、STEM、写作及医学、法律等高精度领域能力增强。核心升级在于个性化功能,能有效利用用户保存的记忆、过往聊天、文件和Gmail上下文,并展示影响回复的记忆来源。该模型将在未来两天内向所有用户推出,个性化改进优先面向网页版Plus和Pro用户,移动版随后跟进;API版本为gpt-5.5-chat-latest。官方表示,升级后的模型能提供更智能、清晰、个性化的答案,语气温暖自然且更简洁。
GPT-5.5 Instant is starting to roll out in ChatGPT. It's a big upgrade, giving you smarter, clearer, and more personaliz...
Introducing Realtime TTS-2, a new generation of voice model built for realtime conversation. It is the first voice model...
SubQ模型发布,这是首个基于完全次二次稀疏注意力架构(SSA)的前沿LLM,拥有1200万token的上下文窗口。它在处理100万token时比FlashAttention快52倍,成本低于Opus的5%。该模型突破传统Transformer注意力计算所有token关系的限制,通过稀疏注意力选择性聚焦重要关系,使长上下文处理的计算量减少近1000倍,显著改变了LLM的成本曲线和扩展方式。
Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse...
Exciting news: UNI-1.1-Max and UNI-1.1 debuts making @LumaLabsAI the #3 lab in the Image Arena across both Text-to-Image...
A new anonymous model debuts at #8 in the Artificial Analysis Text to Image Arena! Peanut's weights are expected to be r...
Grok 4.3 just became the smartest AI in the world at law and money It took #1 on TWO brutal private tests no other model...
Bach-1.0 Preview from Video Rebirth debuts at #6 on the Artificial Analysis Text to Video Leaderboard (No Audio)! Bach-1...
Bach-1.0 Preview from Video Rebirth debuts at #6 on the Artificial Analysis Text to Video Leaderboard (No Audio)! Bach-1...
据传谷歌将在I/O大会发布名为“Omni”的新模型,旨在将Gemini从聊天助手升级为集成写作、图片、视频、长上下文记忆与复杂任务流的全模态生产力平台。该模型可能原生支持视频生成与输出,超越现有的Veo 3.1。同时,Gemini 3.2/3.5版本或专注于提升推理速度与效率,而Ultra版本则向长上下文、重记忆及多步骤工作流方向深化。若消息属实,Gemini将成为首个具备视频输出能力的顶级Omni模型。
GOOGLE I/O 🚨: A NEW OMNI MODEL IS BEING TESTED ON GEMINI FOR VIDEO GENERATION! > "Start with an idea or try a template....
GOOGLE I/O 🚨: A NEW OMNI MODEL IS BEING TESTED ON GEMINI FOR VIDEO GENERATION! > "Start with an idea or try a template....
谷歌正在其Gemini平台测试一款名为“Omni”的新模型,专注于视频生成功能。泄露信息显示,该模型的界面提示用户“从一个想法开始或尝试一个模板”,并注明“由Omni驱动”。这一模型可能与内部代号“Toucan”的视频生成工具密切相关,后者目前由Veo驱动。如果谷歌计划正式发布用于视频生成的Gemini Omni,其性能很可能超越当前的Veo 3.1版本。若消息属实,Gemini将成为首个具备视频输出能力的顶级Omni模型,这标志着谷歌在视频生成领域的重大技术进展,并可能为未来的Google I/O 2026活动预热。此举显示了谷歌在人工智能视频生成方面的持续创新和竞争态势。
🚨 Google updated Gemini 3 Flash in arena It still has the same name "Gemini 3 Flash". However, output quality is two ti...
🚨 Google updated Gemini 3 Flash in arena It still has the same name "Gemini 3 Flash". However, output quality is two ti...
Grok 4.3 release > #1 in caselaw > #1 in corpfin > impressive given significantly lower cost per 1m tokens (5-10x less t...