搜索“OpenAI”

te-of-the-art expressiveness and instruction following for 35+ languages and 50+ dialects. At $13 per 1M characters, Voxtral Speech 3 costs less to run than comparable models from OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. The speech model implicitly interprets the semantic meaning of context better, producing natural speech across a diverse range of emotions and languages. New and improved features include voice cloning from 30+ minutes of sampled audio, custom prompt instructio

匹配字段正文

玩意儿嘛，迭代实在有点太快了，就算买的只是包月。 白买Pro 10X，我好心疼我的钱包。 而且现在C端用户真的很惨啊o(╥﹏╥)o。 又是5小时限额，又是半夜Reset背刺，动不动还要提防后台的降智偷袭。。。 买API，充个几十块钱，哪个拉胯了随时拜拜，简直不要太舒服。 泪目了家人们。 不知道大家有没有相同感受，说实话，这一个月来我是有点憋屈得慌的。 A社、OpenAI、谷歌疯狂发布旗舰模型，反观国产阵营这边，啥动静也没有啊。 不少大佬也在说，LLM这一战基本结束了，硅谷靠Scaling杀死了一切。 但如今，拐点，似乎正在出现。 短短两天，两款国产Flash模型接连登场。 旗舰级的能力，正在以Flash的速度和价格加速普惠。 确实，SOTA这场仗，不是那么好打的。 但当「性价比」这一前提得到满足，另一场更重要的仗—— 大模型产业落地，或许，才刚刚开始。 而这，一直是中国公司，最擅长做的事。 本文来自微信公众号 “量子位”（ID：QbitAI），作者：关注前沿科技，36氪经授权发布。 该文观点仅代表作者本人，36氪平台仅提供信息存储空间服务。+123 好文章，需要你的鼓励量子位特邀作者5

匹配字段正文

型的结果是在我们的评估设置下复现的，可能因其提示词、解码设置、模型版本和评估流程的差异而与官方报告有所不同。 🧭 推理选项 vLLM-Omni SGLang Transformers 最适合 服务与语音交互 高吞吐文本服务 本地评估与研究 输入 音频 + 文本指令 音频 + 文本指令 音频 + 文本指令 输出 文本或文本+语音 文本 文本或文本+语音 接口 OpenAI 兼容 API OpenAI 兼容 API Hugging Face AutoClass 依赖项 requirements-vllm-omni.txt requirements-sglang.txt requirements-transformers.txt 🚀 快速开始 1️⃣ 下载模型 pip install -U huggingface hub hf download inclusionAI/Ming-3.0-audio-flash \ --local-dir inclusionAI/Ming-3.0-audio-flash export MODEL PATH="$PWD/inclusionAI/Ming-3.0-

匹配字段正文

的空间监督数据，强化模型的目标识别、空间定位与交互锚定能力，探索以数据Scaling驱动具身基模型能力跃升的新路径，为提升通用具身交互能力提供关键支撑。 依托约百万条高质量、可验证的空间锚定训练样本，ACE-Brain-0.6实现具身空间感知能力新突破：在八项Grounding评测（七项国际权威公开基准及大晓自建RobotGrounding）中，六项成绩超越OpenAI GPT-6 Astra；七项公开基准平均得分达74.53分，领先Astra的73.68分；在真实机器人场景评测RobotGrounding中取得94.13分，超越Astra的91.10分。 国新办新闻发布会 介绍2026年前三季度工业和信息化发展情况 498人参与 预 告 直播丨国新办就国家应急管理体系和能力现代化建设举行新闻发布会 2026-10-21 16:00 预 告 直播丨国新办就“法治中国建设”举行中外记者见面会 2026-10-21 14:00 预 告 直播丨国新办就中国积极应对气候变化有关情况举行新闻发布会 2026-10-21 10:00 热点 1 一喷去黄提亮、改善皮肤屏障，面部喷雾宣称的“神奇功效”

匹配字段正文

创投发布LP源计划VClubVClub投资机构库投资机构职位推介投资人认证投资人服务 活动搜索项目推荐我要入驻城市合作RSI三块关键拼图，MemT 1.0从模型层补上“记忆”量子位·2026年10月20日 15:28把记忆原生焊进Transformer AI圈，最近集体盯上了三个字母：RSI。 Hinton的RSI论文，22位署名作者里，坐着Bengio、OpenAI首席科学家Pachocki、Anthropic联创Jack Clark……几乎AI的半壁江山都在这儿。 一次改进，如何成为下一次的起点？ 有效的方法需要被保留，走过的弯路也需要被识别，承接这件事的，其实是记忆。 记忆张量在这方面的探索，则给出了一份截然不同的答卷——记忆原生。 何为记忆原生？就是将经验的保留、更新和调用，做成模型可以学习的能力。 当记忆在模型中生根发芽，便构成了记忆原生模型。 其最新开源发布的MemT 1.0系列，就给出了一条颇为具体的路径： 它把经验编码成内部动态状态，再通过训练学习怎样写入、更新和读取，最后让这些状态经由记忆分支参与后续计算。 与外部记忆相比，它不仅能力没打折，未来拓展的天花板也会更高

匹配字段正文

. Computer use is where the model looks weakest. It scores 61.7 on OSWorld V2, which MiniMax says leads Chinese models, but VentureBeat notes it trails the strongest Anthropic and OpenAI models. MiniMax also reports 82 percent on CyberGym and says its internal security experts rated the model strong at code auditing, patching, penetration testing, malware analysis, and reverse engineering. The launch post doesn’t say how those evaluations were run, so treat the security claims as MiniMax’s own u

匹配字段正文

(LLM) with 2.8 trillion parameters. The development marks a move to secure independent AI models and training technologies, shifting away from its previous strategy of relying on OpenAI models. The new model uses a mixture-of-experts (MoE) architecture across all layers, with 2.8 trillion total parameters and 80 billion active parameters. MoE activates only selected expert networks depending on the input tokens. Its predecessor, MAI-Base-1, combined dense feed-forward networks with MoE layers.

匹配字段正文

n Models and Software, Kari Briski, wrote in the blog post announcing these models. Nvidia claims that the Super VL completes tasks faster than competing models from Anthropic and OpenAI (Opus 5, GPT 6.1 Sol, Luna, Astra) when post-trained on specific agentic tasks. These tasks include document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience. While such claims are standard whenever a major tech company announces a new model

匹配字段正文

h higher and cheaper at max effort. How do you deploy MiniMax M3.1 Flash? The model is live on platform.minimax.io as MiniMax-M3.1-Flash. MiniMax supports Anthropic-compatible and OpenAI-compatible access, including Chat Completions and Responses. It is also available through M Plan and the MiniMax Code desktop app. Pricing is a flat $0.15 input and $0.60 output per 1M tokens. Cache reads cost $0.015 per 1M. There is no long-context surcharge. MiniMax M3 costs $0.30/$1.20 up to 512K input and do

匹配字段正文

A separate Fetch API converts a single web page to Markdown. Both cost 75 cents per 1,000 successful requests, and calls made through DeepSeek Harness, Anthropic’s Claude Code or OpenAI Group PBC’s Codex are free until Nov. 1. The releases follow a nearly five-hour API outage on Oct. 15 that Yicai Global counted among DeepSeek’s worst this year. The company apologized afterward, and users received credits pegged to what they had spent the week before. DeepSeek also said it would remove single p

匹配字段正文

their GPUs,” he said. At the same time, some of Nvidia’s customers boast their own AI chips. For instance, Google offers its TPUs, Amazon offers Trainium AI accelerator chips, and OpenAI is developing its own AI chip, Jalapeño. Related:Musing about Muse’s new AI models, unruly DeepSeek “You could see this as an [industry-wide] trend toward integration,” Harrowell said, referring to the idea that as Google, Amazon and OpenAI develop their own chips, they can integrate those chips and optimize the

匹配字段正文

们刚玩完的一款桌游，并询问一个前沿模型谁在获胜。它答不上来。桌游“每一种都有自己的视觉语言，都有自己的图像符号和象征，”他告诉TechCrunch。“对成年人来说，仅凭极少的文字就很容易理解发生了什么，但这些模型无法凭直觉理解这些符号的含义。” “人类可以看着没有任何文字的宜家说明书，组装宜家家具，对吧？为什么AI模型不行？”他说。 这种方法并非完全新颖——OpenAI在2025年为其o系列模型在推理过程中加入了裁剪和缩放功能——但Elorian的卖点在于，它以低成本围绕这一能力构建了整个模型。该公司表示，一个小团队用几百块GPU训练了Glimpse，而且该模型的规模远小于它所对标的那些前沿系统。 Glimpse的定价为每百万输入token 0.60美元、每百万输出token 3美元——比Sonnet和Sol等中端模型便宜，但比Haiku和Luna等轻量级模型贵。 但根据Elorian的评估，Glimpse在科学图表推理和短视频描述方面甚至超过了GPT-6 Astra，在其他视觉推理任务上也具有竞争力。 Elorian在六个月前结束隐身状态，当时完成了一轮由Striker Ventur

匹配字段正文

理”和“复杂软件工程”，正奋力追赶中国竞争对手 Meta于周四发布了一款新的开放权重AI模型，并声称该模型可与领先中国开发者的同类产品相媲美，在某些情况下甚至表现更优。 Muse Spark Open“是迄今为止能力最强的美国开放权重模型”，Meta的AI负责人Alexandr Wang在X上写道。 与所谓的封闭模型——例如Anthropic的Claude或OpenAI的ChatGPT——不同，开发者可以下载开放权重模型的参数（即权重和偏置）并对其进行修改。 过去几年里，中国开发者已成为AI模型开放权重市场的主导者。 阿里巴巴的Qwen、月之暗面的Kimi K3以及Z.ai的GLM模型一直被视为全球开放权重领域的领跑者，但Meta声称其新的开放模型如今已能在相近水平上竞争。 “专为编码和智能体工作打造”，Muse Spark Open“擅长长周期任务、复杂软件工程和多步推理”，Meta在一篇博客文章中表示，并补充说新模型已内置“强有力的安全防护措施”。 Meta表示，Muse Spark Open还针对苹果先进的M5 Ultra芯片进行了“优化”，这标志着两家科技公司之间不同寻常的合作

匹配字段正文

样每块 GPU 持有 78 GB 权重。若用 8 块 GPU，每块将持有 139 GB，这在 141 GB 的 H200 上不留任何空间。设备数量必须能整除专家数量 256。 vLLM Beam 在 vLLM 中提供 day-0 支持。下面的命令通过张量并行在 16 块 GPU（两个 8x H200 节点）上服务 BF16 checkpoint，并暴露一个 OpenAI 兼容 API： 为了更快的生成，添加 DFlash2 投机解码。一个小型草稿模型最多提前提出 15 个 token，Beam 在一次前向传播中检查它们全部，因此每一步可以接受多个 token。代价是草稿模型需要少量额外内存： 这两个命令都会启动一个 OpenAI 兼容服务器，因此你可以将任何 OpenAI 客户端指向 http://localhost:8000/v1 。有关更多服务配置，请参阅为 vLLM 提议的 Beam recipe 。 SGLang Beam 在 SGLang 中也提供 day-0 支持。我们建议使用 DFlash2 投机解码来服务它，下面两个命令都会开启该功能。在 Blackwell 上，NVF

匹配字段正文

ercomputer PLUS: Turn a ChatGPT answer into a tool you can use Zach Mink Good morning, AI enthusiasts, and welcome to our 5,678 new readers. Earlier this week, Meta caught up with OpenAI and Anthropic with its frontier model Muse Wonder. Turns out, the comeback story had one more chapter: a large open-weight release with an unlikely partner. The company just dropped Muse Spark Open, calling it the “most capable” open-weight AI model from the U.S., and tuned it with Apple so that two Macs can wor

匹配字段正文

memory goes to KV cache, letting more concurrent agent sessions run on each node and increasing throughput. Muse Spark Open is available now on Baseten Model APIs. The endpoint is OpenAI-compatible, so you can connect your coding agent and point it at Muse Spark Open without changing your harness. We serve NVFP4 weights which are optimized for Blackwell, and tool calling, reasoning output, and the full 1M context window are supported. For teams running agents at scale, we also offer Dedicated In

匹配字段正文

models is self-serving. The more companies run their own open models, the greater the demand for compute hardware, and open models will likely stay optimized for Nvidia for the foreseeable future. Meanwhile, companies like Meta, Google, Anthropic, and OpenAI are building their own chips and optimizing models for those instead. To push the spread of open AI models further, Nvidia recently invested $12.9 billion in Hugging Face , one of the most important platforms for developing and sharing them.

匹配字段正文

start the server: The checkpoint includes the chat template, which we pass explicitly. The launcher configures the Muse reasoning and tool-call parsers. Send a request through the OpenAI-compatible API: The response separates the reasoning trace in reasoning content from the final answer in content . vLLM The Muse Spark Open reference integration for vLLM supports text and image inputs with the NVFP4 checkpoint that require blackwell series GPUs. After installing the reference plugin and its dep

匹配字段正文

s Grok Voice 3.0 compare to GPT-Live-1 and Gemini 3.8 Live? FeatureGrok Voice 3.0Grok Voice 2.0GPT-Live-1 (Astra, medium)Gemini 3.8 Live Extended Thinking DeveloperSpaceXAISpaceXAIOpenAIGoogle ParametersNot disclosedNot disclosedNot disclosedNot disclosed Context lengthNot disclosedNot disclosedNot disclosedNot disclosed WeightsClosed, API onlyClosed, deprecatedClosed, API onlyClosed, API only Input / outputAudio, text / audio, textAudio, text / audio, textAudio, text / audio, textText, image, v

匹配字段正文

nting Nemotron 3.5 Lightning and Nemotron 3 Ultra. For production serving, SGLang provides a high-performance path for continuous batching, prefix reuse, speculative decoding, and OpenAI-compatible APIs. This blog covers the quick start guide to deploy the model with SGLang. TL;DR Architecture: 120B-parameter mixture-of-experts model with 12B active parameters per token. Modalities: Text, images, documents, and video in; text out. Core optimizations: C-RADIOv4 vision encoding, a hybrid Mamba-Tra

匹配字段正文