Noam Brown@polynoamialAI 评分7575OpenAI 下一代模型家族 Astra 的内部版本解决了数学、量子复杂性和理论计算机科学领域的 10 个重大开放问题，包括计算永久式的新电路下界。OpenAI 认为这将是科学推理领域的一大步。
10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!) GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to see what comes next!
MiniMax (official)@MiniMax_AIAI 评分6060SITUATION EXPLAINED: MiniMax released H3 with open weights, and it's already #1 in video editing. • Priced at $7.80 per minute of 2K video, against Seedance 2.0 at $22.45 and Kling 3.0 at $20.16 • It ranks #1 on Artificial Analysis for video editing, #2 for text-to-video behind Gemini Omni Flash, and #3 for image-to-video • H3 takes text, images, video, and audio in a single context and generates 5-15 second clips at native 2K, 24fps, with native stereo audio • MiniMax is releasing the weights under its Community License within days, which would make it the strongest open-weights video model by a wide margin • It launched the same week ByteDance rolled out Seedance 2.5, which now generates 30-second 4K video from a single prompt • Seedance 2.5's face consistency across a full 30 seconds is the visible jump, earlier models drifted into a different face partway through • Voices are still the obvious tell, the video reads as real long before the audio does • Smaller labs building purpose-built video models are now beating frontier labs at this specific task @theojaffee: "This is gonna be one of the biggest examples of things that people are gonna do with these models, historical simulations, basically. Nostalgia videos for zoomers."
SiliconFlow@SiliconFlowAIAI 评分6161DeepSeek-V4-Flash 正式版 API 上线公测，在 Artificial Analysis 智能指数（AII）中获 50 分，较 4 月版提升 10 分，比 V4 Pro 高 6 分，仅比 GPT-5.6 Luna 低 1 分。
OpenAI 首席执行官奥尔特曼本周在美国国会山展示名为 Astra 的全新 AI 系列模型，重点提升长周期任务处理能力。Astra 支持智能体拆分任务，让多个智能体协同工作，解决单个智能体无法处理的复杂问题。奥尔特曼于 7 月 29 日会见美国参议员拉斐尔·沃诺克和伯尼·莫雷诺。
DeepSeek 发布 V4 系列最新模型 DeepSeek-V4-Flash-0731，拥有 3040 亿参数，智能体能力大幅增强。Artificial Analysis 评测显示其表现优于 428B 参数的 MiniMax M3。定价为每百万输入 token 0.14 美元、每百万输出 token 0.27 美元，可能是当前性价比最高的模型。
DeepSeek 于 7 月 31 日发布 DeepSeek-V4-Flash-0731，并将 V4-Flash API 开放公测。该版本沿用 284B 参数 MoE 架构，通过重新后训练实现性能跃升，在智能体基准上全面超越 V4-Pro（预览版）。API 定价为缓存未命中时每百万输入 token $0.14，输出 token $0.28，权重采用 MIT 许可。
Artificial Analysis@ArtificialAnlys精选AI 评分7878推荐理由：DeepSeek 把 V4 Flash 的智能效率往前推了一步，MIT 许可让商业应用毫无顾虑，做轻量级部署的团队可以立刻换上。
Logan Kilpatrick@OfficialLoganKAI 评分4545与 Google DeepMind 机器人团队的一次对话，探讨我们最新的 Gemini Robotics 2 模型、机器人技术进步的轨迹，以及我们下一步的方向 🤖
Artificial Analysis@ArtificialAnlysAI 评分6262
MiniMax (official)@MiniMax_AIAI 评分6868MiniMax H3 is live on OpenArt 🔊 @MiniMax_AI • 2K video with native audio built right in • 5 to 15 second clips • Text, image, and audio to video • Start frame, end frame, and reference image support • Wonder users get unlimited H3 until August 05. Terms & conditions apply. Try it now: http://openart.ai/suite/create-video/minimax-h3
Greg Brockman@gdbAI 评分4646OK, GPT-5.6 Luna is a bit of a beast. Given the 80% price drop today I decided to try it in Datasette Agent, and it's furiously quick and generates all the SQL, HTML and JavaScript (for Datasette Apps) I could possibly want
Elon Musk@elonmuskAI 评分5050BREAKING: Grok 4.5 outperforms GPT-5.6 Terra across almost all shared benchmarks on AskClash, leading in ACB, GPQA, SWE-P, and Atlas while also achieving a higher overall score. 🔥
Google DeepMind 发布 Gemini Robotics 2，称其为迄今最先进的视觉-语言-动作（VLA）模型，可控制从桌面机械臂到全身人形机器人的各类系统。该模型支持全身运动、精细操作及多机器人协同，开发者可通过候补名单申请早期访问。同期发布的 Gemini Robotics ER 2 专为具身推理设计，已上线 Google AI Studio。
Rohan Paul@rohanpaul_aiAI 评分4444Atomic Chat 将 Kimi K3 量化为 1-bit，模型从 2.8T 参数压缩至 590GB（-62%），保留 78.7% 原始质量和完整 1M 上下文。
1-bit Kimi K3 performs at Opus 5 level on 3D physics! We ran our Atomic Chat quant of Kimi K3 locally on 4x B200 against three cloud models and gave them all the same task, to build a giant anvil drop test as a single HTML file with real physics Outputs: K3 1bit (local): 15.8K tokens, $0 API cost Kimi K3 (API): 15.3K tokens, $0.30 API cost Opus 5: 22.8K tokens, $0.77 API cost GPT 5.6: 14.5K tokens, $0.72 API cost All four got the physics right. But only Kimi made a working winch. The drum turns and the chain drags the flat car off the pad. Opus 5 drew the most detail, road markings and sparks on the hit. And you can run a model at this level on your own box now. That still feels insane to us
Thinking Machines 发布开源推理模型 Inkling Small，在 Intelligence Index 上得分 40，仅比更大的 Inkling（41）低 1 分，而总参数量不到后者的三分之一（276B 总量，12B 激活）。
MiniMax (official)@MiniMax_AIAI 评分6363MiniMax H3 is on Leonardo. Most video models give you a clip. H3 gives you the clip and the soundtrack — with your character and voice locked in from refs. Expect: - Commercial-grade quality across ads, e-commerce, gaming & UI - Best value in its class - One lightweight model, fully multimodal Built for one-shot finished content: brand teasers, product videos, fashion films, talking characters
MiniMax (official)@MiniMax_AIAI 评分5353MiniMax H3 by @Hailuo_AI just landed on Magnific Add characters, motion, and the beat. The model handles the rest → Up to 9 images for character and style → 3 videos for motion and camera control → 2K output, 5 to 15 seconds Try it now on Magnific
MiniMax (official)@MiniMax_AIAI 评分4747非常感谢 @runware 从第一天起就与我们同行 🙌 今天为开发者提供更强能力，开放权重即将面向社区推出。 #MiniMaxH3
MiniMax H3 is now on Runware 🎬 - New open-weights flagship video model from MiniMax - Native synced audio, generated in the same pass as the video - Multi-reference conditioning keeps subjects and scene identity consistent across shots - Extend existing clips or audio segments
MiniMax (official)@MiniMax_AIAI 评分5959创作无需牺牲隐私。⚔️ H3 现已上线 @AskVenice，不久后开源权重将把更多控制权交到社区手中。 #MiniMaxH3
MiniMax H3 from @minimax_ai is now available on Venice. Open-weight multimodal video model with native stereo audio on every generation. Available anonymously.
稀宇科技宣布推出 MiniMax H3 通用多模态视频模型，将于北京时间 8 月 3 日 0 点在魔搭社区正式开源。该模型支持对文本、图像、视频、声音的统一理解，可输出原生双声道音视频，最高支持 15s 2K 分辨率。官方称其默认提供 2K 分辨率，每秒价格不到主流模型的 1/3，768P 分辨率下为主流 720P 的 1/2。
Luma@LumaLabsAIAI 评分4040Seedance 2.5 即将登陆 Luma。上线时敬请期待。
MiniMax 正式推出全能多模态生成模型 H3，可联合理解文本、图像、视频和音频，生成最高 2K 分辨率、15 秒时长且带原生立体声的视频。H3 在指令跟随、文字与品牌呈现、V2V 动作迁移上表现突出，2K 下每秒价格低于主流模型三分之一，768p 下低于主流 720p 价格一半。官方计划近日开源模型权重，以支持开源社区并加速硬件兼容。
推荐理由：MiniMax H3 把全模态生成打到了 2K 分辨率且价格不到主流三分之一，还计划开源权重，这对闭源视频模型是实打实的压力，开发者可以重点关注。
Artificial Analysis@ArtificialAnlysAI 评分7676DeepSeek V4 Flash 0731 在 Artificial Analysis 智能指数上得 50 分，较 4 月版提升 10 分，领先 V4 Pro 6 分，并进入智能与成本帕累托前沿。
Rohan Paul@rohanpaul_aiAI 评分5454🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the configuration details in our official API docs: https://api-docs.deepseek.com/quick_start/agent_integrations/codex
谷歌 Gemini 3.5 Pro 今日短暂出现在 Arena（原 LMArena）平台，约 30 分钟后被移除，有用户仅完成一次 SVG 测试。该模型于 2026 年 5 月 19 日随 Gemini 3.5 系列发布，首发 3.5 Flash，Pro 原计划次月推出但多次延迟，谷歌仅表示正与合作伙伴测试。
Thinking Machines Lab 推出 Inkling-Small，总参数 276B、激活参数 12B，规模约为 Inkling 的 1/4，同样采用混合专家（MoE）架构并支持最高 1M token 上下文。
数字生命卡兹克@Khazix0918AI 评分5151🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the configuration details in our official API docs: https://api-docs.deepseek.com/quick_start/agent_integrations/codex
DeepSeek V4 Flash 正式版在九项 Agent 测试中全部超过 V4 Pro Preview，其中 DeepSWE 长周期功能开发得分提升 7.5 倍。模型结构和尺寸未变，仅重新后训练，价格不变。官方还预告了 Agent 框架 DeepSeek Harness，并计划在 8 号支持接入 codex，提供一键配置脚本，保留 MCP 及项目信任等级。
DeepSeek V4 Flash 正式版发布，模型结构和参数规模不变，仅重做后训练，Terminal Bench 2.1 从 61.8 涨至 82.7，DeepSWE 从 7.3 涨至 54.4。
DeepSeek 今日宣布 DeepSeek-V4-Flash 正式版 API 上线公测，Agent 能力大幅增强，Terminal Bench 2.1 得分 82.7，NL2Repo 54.2，Cybergym 76.7，DeepSWE 54.4，Toolathlon verified 70.3，多项基准远超 V4-Pro-Preview。
火山引擎发布新一代视频生成模型 Seedance 2.5，并将于近期面向企业用户上线 API 服务。徐工集团、小鹏汽车、智元机器人、灵初智能、微分智飞、穹彻智能、Xspark AI（无界智航）等多家企业已达成合作意向，率先接入。Seedance 系列正从内容工具延伸至工业制造、汽车、具身智能等实体产业，用于训练数据合成、场景仿真与作业流程培训。
PolyAI 推出 Dialog-RSN-1，一款直接感知来电者音频而非读取转写的对话模型，将话轮转换、语音识别、函数调用和回复生成融合为单一音频原生模型，已用于实时生产通话。
MiniMax (official)@MiniMax_AIAI 评分4040One take. One ocean. Infinite worlds. @MiniMax_AI H3 is now live on #VidMuse. Longer shots. Smoother motion. Cinematic flow — from the first frame to the last. Sail into a sea of imagination.🌊 #MiniMaxH3
字节跳动今日推出 Seedance 2.5 视频生成模型，单次生成时长从 15 秒提升至 30 秒，并支持多轮延长，可在 30 秒内组织多个具备逻辑关联的镜头。
