<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Orange News · Paglabas ng modelo</title>
    <link>https://new.nat6.net/fil/</link>
    <description>Paglabas ng modelo · Orange News · Subscription sa RSS</description>
    <language>fil</language>
    <lastBuildDate>Sat, 21 Nov 2026 18:45:09 +0000</lastBuildDate>
    <generator>OrangeNews/1.0</generator>
    <ttl>60</ttl>
    <copyright>© 2026 Orange News. Ang copyright ng nilalaman ay pag-aari ng orihinal na may-akda at media; nagbibigay lamang ang site na ito ng pamagat at AI summary at nagli-link pabalik sa pinagmulan.</copyright>
    <image>
      <url>https://new.nat6.net/assets/img/logo-512.png</url>
      <title>Orange News</title>
      <link>https://new.nat6.net/fil/</link>
    </image>
    <atom:link href="https://new.nat6.net/fil/feed/category/ai-models.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>Inilunsad ng Kling AI ang Kling Image 4.0: native na 8K na pagbuo ng larawan at advanced na text rendering</title>
      <link>https://new.nat6.net/fil/news/rhln87mprozseglkairsway74/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/rhln87mprozseglkairsway74/</guid>
      <pubDate>Fri, 20 Nov 2026 16:40:57 +0000</pubDate>
      <description>Inilunsad ng Kling AI ang Kling Image 4.0, na pangunahing tampok ang native na 8K, mas malakas na realismo, pare-parehong grupo ng larawan at advanced na text rendering. Ayon sa pagsusuri ng Artificial Analysis, pumapang-7 ito sa text-to-image at pang-6 sa image editing, na pagtaas ng mahigit 225 E…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://x.com/Kling_ai/status/2123818893122662485">X：可灵 Kling AI (@Kling_ai)</source>
    </item>
    <item>
      <title>Inilabas ng Thinking Machines Lab ang open-weight na modelong Inkling 2 Mini na kayang tumakbo sa laptop</title>
      <link>https://new.nat6.net/fil/news/y7nk77dxae51qwi826gpmvbho/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/y7nk77dxae51qwi826gpmvbho/</guid>
      <pubDate>Fri, 20 Nov 2026 07:00:00 +0000</pubDate>
      <description>Inilabas ng Thinking Machines Lab ang open-weight na modelong Inkling 2 Mini, na gumagamit ng recurrent Transformer architecture at pinapatakbo ng tatlong beses bawat token ang gitnang 12 layer. May kabuuang 26.4B parameter ito, 3.4B ang aktibo, humigit-kumulang 20GB ang memory footprint, at kayang…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Open source ecosystem</category>
      <category>Kaligtasan at alignment</category>
      <source url="https://thinkingmachines.ai/news/inkling-2-mini/">Thinking Machines Lab：News（网页）</source>
    </item>
    <item>
      <title>Tencent Hunyuan naglabas ng Hy4 open weights</title>
      <link>https://new.nat6.net/fil/news/jl6m0eymz9z0kleh54c3guowx/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/jl6m0eymz9z0kleh54c3guowx/</guid>
      <pubDate>Fri, 20 Nov 2026 03:52:40 +0000</pubDate>
      <description>Inilabas ng Tencent Hunyuan ang Hy4 open weights, na nakaposisyon bilang native multimodal flagship MoE model, may kabuuang 770B parameter at 49B aktibong parameter, at sumusuporta sa 1M context. Sabay na inilabas ang Hy4 System Card na may kasamang ulat ng safety assessment, at nagbigay ng mga dow…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <category>Multimodal</category>
      <source url="https://x.com/TencentHunyuan/status/2123625549147070538">X：腾讯混元 (@TencentHunyuan)</source>
    </item>
    <item>
      <title>Gemini 4 Neon, sinubukan: inihambing ni Peter Gostev ang Opus, Fable at iba pang modelo sa unang karanasan</title>
      <link>https://new.nat6.net/fil/news/z0unnn7ryj39a3tuns8b1l4a6/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/z0unnn7ryj39a3tuns8b1l4a6/</guid>
      <pubDate>Thu, 19 Nov 2026 21:28:16 +0000</pubDate>
      <description>Ibinahagi ng Arena ang first-hand na video ng pagsubok sa Gemini 4 Neon matapos itong ilabas, kung saan inihambing ito ni Peter Gostev sa Opus, Fable, Astra at Sol sa mga gawain tulad ng 3D scene, laro, UI at data visualization, at tinasa kung gaano ito kalapit sa frontier.</description>
      <category>Paglabas ng modelo</category>
      <category>Multimodal</category>
      <source url="https://x.com/arena/status/2123528811572846972">X：Arena (@arena)</source>
    </item>
    <item>
      <title>Sinuri ng miyembro ng OpenAI na si Tibo ang GPT-6.2 Astra at binanggit ang mga resulta ng pag-optimize ng AMD at Jalapeño chip</title>
      <link>https://new.nat6.net/fil/news/bnqsw7ispucfily4bdq9vz33j/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/bnqsw7ispucfily4bdq9vz33j/</guid>
      <pubDate>Wed, 18 Nov 2026 22:03:01 +0000</pubDate>
      <description>Binanggit ng miyembro ng OpenAI na si Tibo ang mga resulta ng koponan at sinuri ang GPT-6.2 Astra bilang mahusay sa napakahirap na mga gawain, habang inamin na nakatanggap ng feedback sa istilo nito sa komunikasyon at patuloy na nagpapabuti. Sinasabi ng binanggit na nilalaman na, sa pakikipagtulung…</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://x.com/thsottiaux/status/2123175166519992652">X：Tibo (@thsottiaux)</source>
    </item>
    <item>
      <title>Bukas na ang Gemini 4 Neon sa mga subscriber ng Google AI Pro at Ultra</title>
      <link>https://new.nat6.net/fil/news/dpmse0g0euqubtr0704x8bqvw/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/dpmse0g0euqubtr0704x8bqvw/</guid>
      <pubDate>Wed, 18 Nov 2026 21:37:20 +0000</pubDate>
      <description>Inanunsyo ng Google na mula ngayon, bukas na ang Gemini 4 Neon sa mga subscriber ng Google AI Pro at Ultra, saklaw ang mobile, desktop at web. Ayon sa opisyal, mahusay ang modelong ito sa mga gawaing nangangailangan ng frontier-level na pangangatwiran at mabilis na tugon.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://x.com/GeminiApp/status/2123168704662012079">X：Gemini (@GeminiApp)</source>
    </item>
    <item>
      <title>Inilabas ng Google DeepMind ang Gemini 4 Neon</title>
      <link>https://new.nat6.net/fil/news/lp0snt9lfrx5ju7mg3wma6hnw/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/lp0snt9lfrx5ju7mg3wma6hnw/</guid>
      <pubDate>Wed, 18 Nov 2026 21:35:43 +0000</pubDate>
      <description>Inilabas ng Google DeepMind ang Gemini 4 Neon, na nakatuon sa frontier intelligence at mabilis na end-to-end na pagkumpleto ng gawain, na sumasaklaw sa software engineering, knowledge work at scientific workflow. Nakakuha ang modelo ng 61 puntos sa Artificial Analysis Intelligence Index at pumangat…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Kaligtasan at alignment</category>
      <source url="https://deepmind.google/blog/introducing-gemini-4-neon/">Google DeepMind：Blog（RSS）</source>
    </item>
    <item>
      <title>Inilabas ng OpenAI ang GPT-6.2 Astra at Sol: mas mura ngunit hati ang reaksyon, hawak pa rin ng Anthropic ang pinakamalakas na modelo</title>
      <link>https://new.nat6.net/fil/news/oisr0dfay736le62g425urn8l/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/oisr0dfay736le62g425urn8l/</guid>
      <pubDate>Wed, 18 Nov 2026 10:19:04 +0000</pubDate>
      <description>Binubuod ng Latent Space ang paglabas ng OpenAI ng GPT-6.2 Astra at Sol: nasukat ng Artificial Analysis na naabot ng Sol ang Astra ng Setyembre sa halagang humigit-kumulang 15% ng gastos kada gawain, habang ang bagong Astra ay pumapangalawa sa Intelligence Index na may 62 puntos, kasunod lamang ng…</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Anthropic / Claude</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://latent.space/p/ainews-gpt-6-2-astra-the-2-frontier-model-is-better-and-20-600-cheaper-but-has-mixed-reviews-6-2-sol/">Latent Space（RSS）</source>
    </item>
    <item>
      <title>Tencent Hunyuan Hy4 flash, open weights, may FP8 at Base na bersyon</title>
      <link>https://new.nat6.net/fil/news/ei7hom06r7wgtedvkl2eqbofp/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/ei7hom06r7wgtedvkl2eqbofp/</guid>
      <pubDate>Wed, 18 Nov 2026 04:39:34 +0000</pubDate>
      <description>Inanunsyo ng Tencent Hunyuan na open weights ang Hy4 flash, na may dalawang bersyon: Hy4 flash-FP8 at Hy4 flash-Base. Ang modelo ay may 224B na parameter (12B ang aktibo), sumusuporta sa 1M na konteksto, native multimodal, at nakatuon sa mahusay na pang-araw-araw na produktibidad. Inilabas ang mode…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <category>Multimodal</category>
      <source url="https://x.com/TencentHunyuan/status/2122912575302119554">X：腾讯混元 (@TencentHunyuan)</source>
    </item>
    <item>
      <title>Malalim na pagsusuri sa GPT-6.2 Astra: sobrang matipid ngunit lubhang kulang sa ilang aspeto, mahina pa rin ang komunikasyon at pagiging madaling basahin</title>
      <link>https://new.nat6.net/fil/news/n125mhueen4ul4elntsa1kk1v/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/n125mhueen4ul4elntsa1kk1v/</guid>
      <pubDate>Tue, 17 Nov 2026 22:36:45 +0000</pubDate>
      <description>Inilabas ng OpenAI ang GPT-6.2 Astra at Sol, na sabay nang available sa ChatGPT, Codex at API.</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://mp.weixin.qq.com/s/0cEoTueGjSGtm0-glpYGcA">公众号：数字生命卡兹克</source>
    </item>
    <item>
      <title>Inilunsad ng Perplexity ang GPT-6.2 Astra at GPT-6.2 Sol, inihayag ang WANDR benchmark na resulta</title>
      <link>https://new.nat6.net/fil/news/zy931jr86bk2nk5cv8ww6seah/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/zy931jr86bk2nk5cv8ww6seah/</guid>
      <pubDate>Tue, 17 Nov 2026 22:10:02 +0000</pubDate>
      <description>Inanunsyo ng Perplexity na ang GPT-6.2 Astra at GPT-6.2 Sol ay magagamit na sa Perplexity Search at Computer. Sa kanilang WANDR benchmark, ang GPT-6.2 Astra ay nakakuha ng 0.783, humigit-kumulang $12.54 bawat gawain, mas mataas ng 0.095 kaysa GPT-6 Astra; ang GPT-6.2 Sol ay nakakuha ng 0.746, humig…</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <source url="https://x.com/perplexity_ai/status/2122814545639223348">X：Perplexity (@perplexity_ai)</source>
    </item>
    <item>
      <title>OpenAI 发布 GPT-6.2 Astra 与 Sol：Astra API 每百万 token 输入 8 美元、输出 40 美元，谁能用？</title>
      <link>https://new.nat6.net/fil/news/custom-83051591a31445f7/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/custom-83051591a31445f7/</guid>
      <pubDate>Tue, 17 Nov 2026 21:34:29 +0000</pubDate>
      <description>OpenAI 已发布 GPT-6.2 Astra 和 GPT-6.2 Sol。Astra 面向更复杂的专业任务，Sol 面向日常工作。官方称，两款模型都能用更少的输出 token 取得更好结果。开发者在 API 里要找的名字是 gpt-6.2-astra 和 gpt-6.2-sol。 以上事实核自 OpenAI官方介绍。官方系统卡标注日期为 2026 年 11 月 17 日，并写 “Today, we are releasing”。本页核验时间是 2026 年 11 月 18 日（北京时间）。 谁能用，从哪用？ 官方介绍写明：即日起，Plus、Pro、Enterprise、Edu 用户可在…</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Paglabas ng mga modelo</category>
      <category>Kaligtasan at alignment</category>
      <source url="https://openai.com/index/introducing-gpt-6-2-astra-sol/">OpenAI</source>
    </item>
    <item>
      <title>Inilabas ng Microsoft AI ang unang speech-to-speech model na MAI Realtime 1, $0.01 kada minutong audio ang presyo</title>
      <link>https://new.nat6.net/fil/news/eewx1bbnzfmbj6r40v7b6s6nb/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/eewx1bbnzfmbj6r40v7b6s6nb/</guid>
      <pubDate>Tue, 17 Nov 2026 18:16:17 +0000</pubDate>
      <description>Inilabas ng Microsoft AI ang MAI Realtime 1, ang kauna-unahang speech-to-speech model nito, na may panimulang presyong $0.01 kada minutong audio at sinasabing mas mahusay kaysa GPT Realtime 2.1 sa Big Bench Audio at τ-Voice benchmarks.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://microsoft.ai/news/introducing-mai-realtime-1/">Microsoft AI：官方博客（RSS）</source>
    </item>
    <item>
      <title>Naglabas ang Microsoft Ignite 2026 ng 7 bagong modelo sa seryeng MAI</title>
      <link>https://new.nat6.net/fil/news/z9zhvfj8o8zp2yp433eizf3uk/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/z9zhvfj8o8zp2yp433eizf3uk/</guid>
      <pubDate>Tue, 17 Nov 2026 18:16:05 +0000</pubDate>
      <description>Naglabas ang Microsoft sa Ignite 2026 ng 7 bagong modelo sa seryeng MAI, na may puntong cost-effectiveness.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://microsoft.ai/news/microsoft-ignite-2026/">Microsoft AI：官方博客（RSS）</source>
    </item>
    <item>
      <title>Inilabas ng Microsoft AI ang MAI Code 2 at Code 2 Flash na mga coding model</title>
      <link>https://new.nat6.net/fil/news/c3qj65urfo345mc80xz5jdefb/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/c3qj65urfo345mc80xz5jdefb/</guid>
      <pubDate>Tue, 17 Nov 2026 18:15:16 +0000</pubDate>
      <description>Inilabas ng Microsoft AI ang dalawang modelong MAI Code 2 at MAI Code 2 Flash na nakatuon sa pang-araw-araw na coding sa negosyo, na may puntong cost-effectiveness. Ayon sa teksto, sa average ng apat na coding benchmark, naabot ng MAI Code 2 ang 98% ng performance ng GPT-6.1 Sol sa 40% ng gastos, a…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <source url="https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/">Microsoft AI：官方博客（RSS）</source>
    </item>
    <item>
      <title>Inilabas ng OpenAI ang GPT-6.2 Astra at GPT-6.2 Sol</title>
      <link>https://new.nat6.net/fil/news/jr7fqw95e5ihuevr1rzib208l/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/jr7fqw95e5ihuevr1rzib208l/</guid>
      <pubDate>Tue, 17 Nov 2026 17:45:00 +0000</pubDate>
      <description>Inilabas ng OpenAI ang GPT-6.2 Astra para sa komplikadong propesyonal na trabaho at ang GPT-6.2 Sol para sa pang-araw-araw na trabaho, na parehong nakakakuha ng mas magandang resulta sa mas kaunting output token.</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://openai.com/index/introducing-gpt-6-2-astra-sol/">OpenAI：官网动态（RSS · 排除企业/客户案例）</source>
    </item>
    <item>
      <title>Artificial Analysis, sinuri ang GPT-6.2 Astra at Sol: tumaas sa ikalawang puwesto ang intelligence index</title>
      <link>https://new.nat6.net/fil/news/m4ti1m7rs04pdy8067c1k0uas/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/m4ti1m7rs04pdy8067c1k0uas/</guid>
      <pubDate>Mon, 16 Nov 2026 16:00:00 +0000</pubDate>
      <description>Sinuri ng Artificial Analysis ang GPT-6.2 Astra at GPT-6.2 Sol na inilabas ng OpenAI; kapwa sila nasa ikalawang puwesto sa Artificial Analysis Intelligence Index, kasunod lamang ng Claude Fable 5.2.</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <source url="https://artificialanalysis.ai/articles/openai-has-the-2-model-in-the-intelligence-index-with-the-release-of-gpt-6-2-astra-and-gpt-6-2-sol">Artificial Analysis 完整文章（网页）</source>
    </item>
    <item>
      <title>Inilabas ng Google ang Gemini 4 Carbon at Hydrogen, kasama ang Decisions API at self-hosted agent sandbox</title>
      <link>https://new.nat6.net/fil/news/n3qi466u1gbxj2i8q7g5snhr0/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/n3qi466u1gbxj2i8q7g5snhr0/</guid>
      <pubDate>Mon, 16 Nov 2026 15:57:52 +0000</pubDate>
      <description>Inilabas ng Google ang Gemini 4 Carbon (gemini-4-carbon, 1M shared token pool, $2.00/$10.00 bawat 1M tokens bago ang Nobyembre 30.</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>Multimodal</category>
      <source url="https://x.com/_philschmid/status/2122358498461691913">X：Philipp Schmid（Google DeepMind 开发者体验） (@_philschmid)</source>
    </item>
    <item>
      <title>Tencent Hunyuan inilabas ang Hy4 flash: 224B kabuuang parameter, 1M context na pang-araw-araw na matipid na modelo</title>
      <link>https://new.nat6.net/fil/news/u36vxmxeopbsx4if4lap30899/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/u36vxmxeopbsx4if4lap30899/</guid>
      <pubDate>Mon, 16 Nov 2026 04:29:37 +0000</pubDate>
      <description>Inilunsad ng Tencent Hunyuan ang Hy4 flash, na may 224B kabuuang parameter (kabilang ang 64B Engram parameter), 12B aktibong parameter, sumusuporta sa multimodal at 1M context, at malaki ang pagbaba ng presyo ng API kumpara sa Hy4 (0.5 yuan input, 2 yuan output kada milyong tokens).</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Multimodal</category>
      <source url="https://hunyuan.tencent.com/research/100145?langVersion=zh">腾讯混元：Research（API）</source>
    </item>
    <item>
      <title>Inilabas ng Tongyi Qianwen ang Qwen4 Max, Intelligence Index 9 puntos na mas mataas kaysa Qwen3.8 Max at mas mababa ang gastos</title>
      <link>https://new.nat6.net/fil/news/r9e7azie93xzxxdf6yyd1jmqe/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/r9e7azie93xzxxdf6yyd1jmqe/</guid>
      <pubDate>Mon, 16 Nov 2026 03:46:34 +0000</pubDate>
      <description>Inanunsyo ng Tongyi Qianwen ang paglabas ng Qwen4 Max; ayon sa datos ng Artificial Analysis, nakakuha ito ng 43 sa Intelligence Index, 9 puntos na mas mataas kaysa Qwen3.8 Max (xhigh), at 3 puntos na nahuhuli sa Kimi K3.1 Pro. $3.27 ang gastos kada gawain, mga isang-katlo ng Qwen3.8 Max, at 39% ang…</description>
      <category>Paglabas ng modelo</category>
      <category>Multimodal</category>
      <source url="https://x.com/Alibaba_Qwen/status/2122174463479542126">X：通义千问 / Qwen (@Alibaba_Qwen)</source>
    </item>
    <item>
      <title>Qwen4-Max 登顶 Agent Arena 中国模型榜首，Alibaba 排名升至总榜第14</title>
      <link>https://new.nat6.net/fil/news/kzt8bjcin243qefkuiq49c2i6/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/kzt8bjcin243qefkuiq49c2i6/</guid>
      <pubDate>Mon, 16 Nov 2026 02:21:45 +0000</pubDate>
      <description>Arena 榜单显示 Qwen4-Max 以 +7.5% 净提升在 Agent Arena 总排名第14，成为第一中国模型，使 Alibaba 成为排名第一的中国实验室、总榜第5，仅次于 Anthropic、Google、OpenAI 和 Meta。</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <source url="https://x.com/arena/status/2122153116959482051">X：Arena (@arena)</source>
    </item>
    <item>
      <title>Qwen 发布 Qwen4-Max：2.4T 参数骨干、百万 token 上下文与长程智能体能力</title>
      <link>https://new.nat6.net/fil/news/nqcfrkilrh7nhnpd03t9q1zfh/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/nqcfrkilrh7nhnpd03t9q1zfh/</guid>
      <pubDate>Mon, 16 Nov 2026 02:00:00 +0000</pubDate>
      <description>Qwen 团队发布 Qwen4-Max，基于新 Qwen4 架构，拥有 2.4T 骨干参数、每 token 激活 71B、328B N-gram Embedding 参数，具备原生视觉能力和 100 万 token 上下文窗口。</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://qwen.ai/blog?id=qwen4-max">Qwen：Blog Retrieval（API）</source>
    </item>
    <item>
      <title>Tencent Hunyuan naglabas ng real-time voice model na Hy Realtime, may bagong tool calling at kakayahang kumanta</title>
      <link>https://new.nat6.net/fil/news/qp41q77z9ge5hsyrh9kra0c7k/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/qp41q77z9ge5hsyrh9kra0c7k/</guid>
      <pubDate>Fri, 13 Nov 2026 08:32:39 +0000</pubDate>
      <description>Naglabas ang Tencent Hunyuan ng bagong real-time voice model na Hy Realtime, na may bagong Agentic tool calling at kakayahang kumanta bukod sa mga batayang kakayahan tulad ng voice understanding, pagsunod sa instruksyon, at pagpapahayag na parang tao. Nailunsad na ito sa voice call ng Yuanbao at sa…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://mp.weixin.qq.com/s/bFxAT04ohJs9hPcjjin3bg">公众号：腾讯混元</source>
    </item>
    <item>
      <title>Meituan LongCat, inilabas ang open source na LongCat-Video-Avatar 1.6 na audio-driven digital human video generation model</title>
      <link>https://new.nat6.net/fil/news/habommgck5ql3tdf688ca370s/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/habommgck5ql3tdf688ca370s/</guid>
      <pubDate>Fri, 13 Nov 2026 02:45:33 +0000</pubDate>
      <description>Inilabas ng Meituan LongCat team ang open source na audio-driven human video generation model na LongCat-Video-Avatar 1.6, na nagpabuti sa naturalness ng bibig, lip-sync, koordinasyon ng galaw ng mukha at katawan, at visual consistency ng mahabang video, na nagpapagaan sa pag-ipon ng error sa autor…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <source url="https://huggingface.co/meituan-longcat/LongCat-Video-Avatar-1.6">美团 LongCat：HuggingFace 新模型</source>
    </item>
    <item>
      <title>Gemini 4 Hydrogen, ipapamahagi sa lahat ng user sa buong mundo simula ngayon</title>
      <link>https://new.nat6.net/fil/news/zlbngirpqifhhi3bd5pbmixu4/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/zlbngirpqifhhi3bd5pbmixu4/</guid>
      <pubDate>Thu, 12 Nov 2026 20:50:50 +0000</pubDate>
      <description>Inanunsyo ng Google Gemini na ang Gemini 4 Hydrogen ay ipapamahagi sa lahat ng user sa buong mundo simula ngayon, saklaw ang desktop, mobile at web. Hindi na kailangang pumili ng modelo ang mga libreng user; kapag ang prompt ay angkop para sa modelong ito, magagamit nila ito sa pamamagitan ng awtom…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://x.com/GeminiApp/status/2120982674638864797">X：Gemini (@GeminiApp)</source>
    </item>
    <item>
      <title>Sinuri ng Artificial Analysis ang Gemini 4 Hydrogen: intelligence index na 54, malapit sa Sonnet 5.5 at halos 32% ng gastos nito</title>
      <link>https://new.nat6.net/fil/news/ajvxvhu8bh001pourbhhjl0ir/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/ajvxvhu8bh001pourbhhjl0ir/</guid>
      <pubDate>Thu, 12 Nov 2026 18:07:16 +0000</pubDate>
      <description>Sinuri ng Artificial Analysis ang bagong inilabas na Gemini 4 Hydrogen ng Google: ang bagong xhigh reasoning tier ay nakakuha ng 54 puntos sa Intelligence Index, isang puntos na mas mababa kaysa sa Claude Sonnet 5.5 (max) at GPT-6.1 Sol (max), at ang gastos kada gawain ay humigit-kumulang 32% ng So…</description>
      <category>Paglabas ng modelo</category>
      <category>Multimodal</category>
      <source url="https://x.com/ArtificialAnlys/status/2120941511223185722">X：Artificial Analysis (@ArtificialAnlys)</source>
    </item>
    <item>
      <title>Inilabas ng Google DeepMind ang Gemini 4 Hydrogen</title>
      <link>https://new.nat6.net/fil/news/uy1ayrrlqx1snqsv4lub7whsa/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/uy1ayrrlqx1snqsv4lub7whsa/</guid>
      <pubDate>Thu, 12 Nov 2026 18:00:50 +0000</pubDate>
      <description>Inilabas ng Google DeepMind ang Gemini 4 Hydrogen, na nakatuon sa mabilis na pagpapatupad ng pang-araw-araw na coding at AI agent workflow, at bukas sa lahat ng antas sa pamamagitan ng Gemini API, AI Studio, Gemini app, Google Search at iba pa.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://deepmind.google/blog/introducing-gemini-4-hydrogen/">Google DeepMind：Blog（RSS）</source>
    </item>
    <item>
      <title>Edge0, naglabas ng open source na 145B na modelong pang-edge na kayang tumakbo nang lokal sa laptop</title>
      <link>https://new.nat6.net/fil/news/icmwp7qazk244f7fn6zlw8882/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/icmwp7qazk244f7fn6zlw8882/</guid>
      <pubDate>Thu, 12 Nov 2026 18:00:00 +0000</pubDate>
      <description>Inanunsyo ng Edge0 ang paglabas ng open source na Edge0-145B at Edge0-120B na mga modelo kasama ang pinag-isang inference framework at deployment tools, at sinabing ang Edge0-145B ang unang 145B-parameter na open source na modelo na kayang tumakbo nang lokal sa personal na computer.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <source url="https://edge0.ai/blog/introducing-edge0-145b-frontier-ai-capabilities-on-your-own-computer">Hacker News：AI 热帖</source>
    </item>
    <item>
      <title>Pagsusuri ng MOSS-Live full-duplex real-time na interaksyon: paghahambing sa GPT-Live-1 at Doubao</title>
      <link>https://new.nat6.net/fil/news/rcjkjkhq7xxwbhh9wnv15fqy1/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/rcjkjkhq7xxwbhh9wnv15fqy1/</guid>
      <pubDate>Thu, 12 Nov 2026 16:00:00 +0000</pubDate>
      <description>Inilabas ng OpenMOSS team ang pagsusuri ng MOSS-Live full-duplex voice interaction evaluation. Batay sa chart ng MOSS official website noong Nobyembre 13, 2026, nangunguna ang MOSS-Live sa anim na full-duplex na indicator sa lahat ng kalahok na modelo, may 2.00% na rate ng maling pagsingit sa usapa…</description>
      <category>Paglabas ng modelo</category>
      <source url="https://openmoss.ai/blog/cn/moss-live/">OpenMOSS / 复旦NLP / 上海创智 / MOSI（网页）</source>
    </item>
    <item>
      <title>Runway, inilabas ang open-weight na 27B world action model na Praxis-1 sa ilalim ng Apache 2.0 license</title>
      <link>https://new.nat6.net/fil/news/bx3sq4dxfqblyztkuiktiz0wv/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/bx3sq4dxfqblyztkuiktiz0wv/</guid>
      <pubDate>Wed, 11 Nov 2026 23:25:00 +0000</pubDate>
      <description>Inanunsyo ng Runway na ang world action model na Praxis-1 ay bukas sa lahat sa ilalim ng Apache 2.0 open weights. Ang modelo ay isang 27B parameter na MoE diffusion Transformer na sabay na naghuhula ng video at aksyon, at nangunguna sa NVIDIA RoboLab-120 leaderboard na may 46.2% success rate, lampa…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <source url="https://runwayml.com/news/company-news/praxis-1-is-now-available">Runway：News（网页）</source>
    </item>
    <item>
      <title>Gemini Decision 1 是干嘛的？不聊天、只做判断题，0.2 秒出结果，输出不收钱</title>
      <link>https://new.nat6.net/fil/news/custom-221eeb31c52349f7/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/custom-221eeb31c52349f7/</guid>
      <pubDate>Wed, 11 Nov 2026 19:10:04 +0000</pubDate>
      <description>一次请求最多带 1000 道判断题，官方称端到端大约 0.2 秒出结果，输出不收费。 2026 年 11 月 11 日，Google 推出 Gemini Decisions API，背后是新模型 Gemini Decision 1，模型名 gemini-decision-1，现在可以在 Google AI Studio 的 Gemini API 里调用。 下面按官方博客和开发者文档来写。速度、价格和评测都是官方说法，本站没有实测。 它和普通聊天模型有什么不一样？ 普通聊天模型是一个字一个字往外写。Gemini Decision 1 不生成回答文本，而是对你给出的选项打分，一次前向计算就返回概…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Mga update sa produkto</category>
      <category>Multimodal</category>
      <source url="https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/">Google</source>
    </item>
    <item>
      <title>MiniMax 放出 H3-Image：4K 图怎么出，许可和入口先看清</title>
      <link>https://new.nat6.net/fil/news/custom-5c5e3b66f469443e/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/custom-5c5e3b66f469443e/</guid>
      <pubDate>Wed, 11 Nov 2026 06:14:17 +0000</pubDate>
      <description>MiniMax 发布了开放权重图像模型 H3-Image，支持文生图、参考图生成和改图，最多 16 张参考图。官方称可直接生成 4096×4096 的 4K 图，并称不到 10 秒（测试硬件未说明）。想先用，可以走海螺 AI、开放平台 API 和 MiniMax Code；权重在 Hugging Face。站内已有一条聚合短讯，这篇只补充入口、许可和官方价目，不重复那条快讯。本站没有跑过实测，下文速度、倍数和价目均为官方公布，不是本站结论。 官方说它能做什么？ H3-Image 建在 H3 之上，文生图和改图做在同一个检查点里。官方模型卡还写：生成和编辑不必换模型；可以直接出大约 16MP…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Multimodal</category>
      <category>Mga tutorial at praktika</category>
      <source url="https://www.minimax.io/blog/minimax-h3-image">MiniMax</source>
    </item>
    <item>
      <title>Inilabas ng MiniMax ang open source na modelong pang-generate ng larawan na H3-Image, nakakagawa ng 4K na larawan sa loob ng 10 segundo</title>
      <link>https://new.nat6.net/fil/news/eel6o8sxfldx5itfa8f8wigc0/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/eel6o8sxfldx5itfa8f8wigc0/</guid>
      <pubDate>Tue, 10 Nov 2026 16:23:20 +0000</pubDate>
      <description>Inilabas ng MiniMax ang H3-Image, isang open weights na image generation model na nakabatay sa H3. Sinusuportahan nito ang text-to-image, reference image generation at image editing, at nakakagawa ng 4096 × 4096 na 4K na larawan, na may bilis na hanggang 5 beses kaysa sa mga pangunahing solusyon.</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <source url="https://www.minimax.io/blog/minimax-h3-image">MiniMax：Blog（网页）</source>
    </item>
    <item>
      <title>Inilabas ng Google DeepMind ang frontier model na Gemini 4 Carbon</title>
      <link>https://new.nat6.net/fil/news/w8cvkd1j2o1c3m8ikh8x2kpfx/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/w8cvkd1j2o1c3m8ikh8x2kpfx/</guid>
      <pubDate>Tue, 10 Nov 2026 16:03:47 +0000</pubDate>
      <description>Pormal na binuksan ng Google DeepMind ang Gemini 4 Carbon para sa mga developer, enterprise, scientist at ordinaryong user, na inilalarawan bilang pinakamalakas nilang frontier model, na nakatuon sa cybersecurity capabilities at long-horizon agent workflows sa complex knowledge work, software engin…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <source url="https://deepmind.google/blog/gemini-4-carbon-is-now-available/">Google DeepMind：Blog（RSS）</source>
    </item>
    <item>
      <title>Inilabas ng Baidu ang ERNIE 5.5 Flash Preview, open source ang weight bago ang ika-25 ng Nobyembre</title>
      <link>https://new.nat6.net/fil/news/qd3f4j2hd0t98tqf28iv7pu2j/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/qd3f4j2hd0t98tqf28iv7pu2j/</guid>
      <pubDate>Tue, 10 Nov 2026 16:00:00 +0000</pubDate>
      <description>Inilabas ng Baidu ang ERNIE 5.5 Flash Preview, ang unang post-training na bersyon batay sa naunang Base model, na may 280B kabuuang parameter, 14B aktibong parameter, at sumusuporta sa 1M na konteksto. Nasa Qianfan platform na ang API ng modelo, at open source ang weight bago ang ika-25 ng Nobyembr…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Multimodal</category>
      <source url="https://ernie.baidu.com/zh/release/ernie-5-5-flash-preview/">百度文心 ERNIE 官方博客</source>
    </item>
    <item>
      <title>Nangunguna ang Anthropic Fable 5.2 sa Remote Labor Index na may 41.3% sa mga totoong freelance na gawain</title>
      <link>https://new.nat6.net/fil/news/oesqqhekbr8e1bekkcitpguqh/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/oesqqhekbr8e1bekkcitpguqh/</guid>
      <pubDate>Tue, 10 Nov 2026 14:07:47 +0000</pubDate>
      <description>Ayon sa ulat ng The Decoder, ang bagong modelong Fable 5.2 ng Anthropic ay niraranggo sa unang pwesto na may 41.3% sa Remote Labor Index benchmark ng Scale Labs at Center for AI Safety, isang pagsubok na sumusukat kung kayang gawin ng AI agent ang mga totoong freelance na proyekto sa kalidad na kat…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Anthropic / Claude</category>
      <source url="https://the-decoder.com/anthropics-fable-5-2-completes-41-percent-of-real-freelance-jobs-up-from-2-5-percent-a-year-ago/">The Decoder：AI News（RSS）</source>
    </item>
    <item>
      <title>Nasubukan ang Claude Fable 5.2: isang beses na direktang output ng video at 3D na web page, namumukod ang aesthetics at detalye</title>
      <link>https://new.nat6.net/fil/news/po2cdt1whxfqyfew3e81l8mb0/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/po2cdt1whxfqyfew3e81l8mb0/</guid>
      <pubDate>Mon, 09 Nov 2026 23:21:03 +0000</pubDate>
      <description>Opisyal na inilabas ng Anthropic ang Claude Fable 5.2 kaninang madaling araw. Agad na sinubukan ng may-akda ang 10 gawain kabilang ang video animation, Three.js 3D scene, at 3D na web page ng Apple Jing'an Store; lahat ay direktang na-output sa isang beses, at walang natuklasang factual error sa mg…</description>
      <category>Paglabas ng modelo</category>
      <category>Anthropic / Claude</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Multimodal</category>
      <source url="https://mp.weixin.qq.com/s/sKoVHL_oNsik9vtUjr0lHg">公众号：数字生命卡兹克</source>
    </item>
    <item>
      <title>Sinuri ng Artificial Analysis ang Claude Fable 5.2: nanguna sa Intelligence Index v5 na may 65 puntos</title>
      <link>https://new.nat6.net/fil/news/m23i0fgxrk1t2gqysu6mqv82s/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/m23i0fgxrk1t2gqysu6mqv82s/</guid>
      <pubDate>Mon, 09 Nov 2026 19:45:12 +0000</pubDate>
      <description>Sinuri ng Artificial Analysis ang Claude Fable 5.2 na inilabas ng Anthropic, na nanguna sa Artificial Analysis Intelligence Index v5 na may 65 puntos, nangunguna sa Gemini 4 Carbon (60) at Claude Opus 5.5 (59).</description>
      <category>Paglabas ng modelo</category>
      <category>Anthropic / Claude</category>
      <source url="https://x.com/ArtificialAnlys/status/2119878995386982508">X：Artificial Analysis (@ArtificialAnlys)</source>
    </item>
    <item>
      <title>Claude Mythos 5.2, hinulaan ang istruktura ng AAV capsid ng gene therapy vector</title>
      <link>https://new.nat6.net/fil/news/rxx1qxfjfuioem95lr868hhwq/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/rxx1qxfjfuioem95lr868hhwq/</guid>
      <pubDate>Mon, 09 Nov 2026 18:56:56 +0000</pubDate>
      <description>Ipinakilala sa opisyal na video ng Claude na hinulaan ng Claude Mythos 5.2 ang kumpletong istruktura ng AAV capsid, ang protein shell na ginagamit sa paghahatid ng gene therapy. Sumulat ang modelo ng code upang patakbuhin ang open source na folding model sa mga katabing pangkat ng limang chain, at…</description>
      <category>Paglabas ng modelo</category>
      <category>Anthropic / Claude</category>
      <category>Paglabas ng mga modelo</category>
      <category>Multimodal</category>
      <source url="https://www.youtube.com/watch?v=FMbXUxn_nBw">Claude：YouTube（RSS）</source>
    </item>
    <item>
      <title>Meituan LongCat-2.5, open source: 1.6 trillion parameter na native multimodal na modelo, MIT license na sumusuporta sa komersyal na paggamit</title>
      <link>https://new.nat6.net/fil/news/x87pxgy2ecovexu57fu5aot9g/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/x87pxgy2ecovexu57fu5aot9g/</guid>
      <pubDate>Mon, 09 Nov 2026 14:00:00 +0000</pubDate>
      <description>Inilabas ng Meituan LongCat ang LongCat-2.5 bilang open source, na may MIT license na sumusuporta sa komersyal na deployment at muling pagsasanay. Ang kabuuang parameter ng modelo ay 1.6 trillion, humigit-kumulang 48 bilyon ang aktibo bawat token, sumusuporta sa isang milyong konteksto, at nakatuon…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <category>Multimodal</category>
      <source url="https://mp.weixin.qq.com/s/EGLDrPxND-xGTyStJIDXrw">公众号：龙猫LongCat（美团）</source>
    </item>
    <item>
      <title>Inilabas ng Meta ang pinakamaliit at pinakamabisang modelong Muse Ember, para sa madalas na gawain ng AI agent</title>
      <link>https://new.nat6.net/fil/news/cox936nwax49rkiimi1oicgzu/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/cox936nwax49rkiimi1oicgzu/</guid>
      <pubDate>Mon, 09 Nov 2026 07:00:00 +0000</pubDate>
      <description>Inilabas ng Meta ang Muse Ember (muse-ember-1.0), na itinuturing na pinakamaliit at pinakamabisang modelo para sa pagsasagawa ng komplikadong pangangatwiran, na nakatuon sa mga karaniwang paulit-ulit na gawain ng AI agent tulad ng pagtawag sa tool, mga subtask sa coding, routing, extraction at pag-…</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Multimodal</category>
      <source url="https://research.meta.ai/blog/muse-ember">Meta AI：Research Blog（网页）</source>
    </item>
    <item>
      <title>Zyphra inilabas ang open source na TTS model na ZONOS3-Mini, kayang tumakbo nang real-time sa browser ng telepono</title>
      <link>https://new.nat6.net/fil/news/r5h0cwwdh1peqbg2rk6k9ljrh/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/r5h0cwwdh1peqbg2rk6k9ljrh/</guid>
      <pubDate>Sun, 08 Nov 2026 16:00:00 +0000</pubDate>
      <description>Inilabas ng Zyphra ang ZONOS3-Mini, isang Apache 2.0 open source na zero-shot voice cloning TTS model na gumagamit ng sparse MoE architecture, may kabuuang 800 milyong parameter at 100 milyong aktibong parameter, may Time To First Audio na mas mababa sa 570 ms sa browser ng Pixel 9 at mas mababa sa…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <category>Open source ecosystem</category>
      <source url="https://www.zyphra.com/our-work/zonos3-mini">Zyphra Research（网页）</source>
    </item>
    <item>
      <title>Inilabas ng Anthropic ang Claude Fable 5.2 at Claude Mythos 5.2</title>
      <link>https://new.nat6.net/fil/news/va9trgwaw3r4por6vdqbjxrhd/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/va9trgwaw3r4por6vdqbjxrhd/</guid>
      <pubDate>Sun, 08 Nov 2026 16:00:00 +0000</pubDate>
      <description>Inilabas ng Anthropic ang Claude Fable 5.2 at Claude Mythos 5.2, na parehong nakabatay sa isang modelong batayan; ang Mythos ay nangangailangan ng aplikasyon para sa access sa pamamagitan ng Cyber o Life Sciences Verification Program.</description>
      <category>Paglabas ng modelo</category>
      <category>Anthropic / Claude</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Multimodal</category>
      <category>Kaligtasan at alignment</category>
      <source url="https://www.anthropic.com/claude-fable-and-mythos-5-2">Anthropic：Newsroom（网页）</source>
    </item>
    <item>
      <title>Inilabas ng ByteDance ang Seedream 5.5 image creation model: sumusuporta sa 100-round editing at native RGBA transparent output</title>
      <link>https://new.nat6.net/fil/news/xsvt2hlwuovul9sdvazcw7cxk/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/xsvt2hlwuovul9sdvazcw7cxk/</guid>
      <pubDate>Sun, 08 Nov 2026 16:00:00 +0000</pubDate>
      <description>Inilabas ng Byte Seed ang bagong henerasyong image creation model na Seedream 5.5, na may tatlong upgrade: pixel-level na pagpapanatili ng mga hindi na-edit na bahagi, suporta para sa hanggang 100 round ng tuloy-tuloy na pag-edit sa isang akda; bagong native RGBA transparent material generation na…</description>
      <category>Paglabas ng modelo</category>
      <category>Paglabas ng mga modelo</category>
      <source url="https://seed.bytedance.com/zh/blog/%E7%94%9F%E6%88%90%E6%9B%B4%E7%9C%9F%E5%AE%9E-%E7%99%BE%E8%BD%AE%E7%BC%96%E8%BE%91%E6%9B%B4%E7%A8%B3%E5%AE%9A-seedream-5-5-%E5%8F%91%E5%B8%83">字节 Seed：Research Feed（网页内嵌数据）</source>
    </item>
    <item>
      <title>Sinubukan ng Artificial Analysis ang pagbuo at pag-edit ng larawan ng GPT Image 2.6</title>
      <link>https://new.nat6.net/fil/news/pqbbgx5sn86iy7v4izf1qv49z/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/pqbbgx5sn86iy7v4izf1qv49z/</guid>
      <pubDate>Fri, 06 Nov 2026 23:40:40 +0000</pubDate>
      <description>Nagsagawa ng benchmark ang Artificial Analysis sa GPT Image 2.6 na inilabas ng OpenAI noong Nobyembre 4, kung saan ang dalawang modelo ay pumanguna at pumangalawa sa AA-Image-T2I v2.1 at AA-Image-Editing v2.1.</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <source url="https://x.com/ArtificialAnlys/status/2118851088111865983">X：Artificial Analysis (@ArtificialAnlys)</source>
    </item>
    <item>
      <title>Inilabas ng Ant Ling ang Ling-3.2-flash: native multimodal, 560B kabuuang parameter, 25B aktibo bawat hakbang</title>
      <link>https://new.nat6.net/fil/news/apnpwouw9phkt81pijcxcl1a1/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/apnpwouw9phkt81pijcxcl1a1/</guid>
      <pubDate>Fri, 06 Nov 2026 11:04:00 +0000</pubDate>
      <description>Inilabas ng Ant Ling ang native multimodal na modelong Ling-3.2-flash, na may 560B kabuuang parameter at 25B aktibong parameter bawat hakbang, sumusuporta sa 1M native na mahabang konteksto, at pinatibay ang kakayahan sa visual feedback self-check at adaptive na pagbangon mula sa anomalya.</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>Multimodal</category>
      <source url="https://mp.weixin.qq.com/s/k1AUP6MNS0zqtcc92idrcg">公众号：蚂蚁百灵（Ling）</source>
    </item>
    <item>
      <title>In-open source ng Xiaomi ang MiMo-V2-Flash: 309B total parameters na MoE model, nakatuon sa reasoning, coding at agents</title>
      <link>https://new.nat6.net/fil/news/cmubjtbhi001srofllk26g3ou/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/cmubjtbhi001srofllk26g3ou/</guid>
      <pubDate>Thu, 05 Nov 2026 20:56:45 +0000</pubDate>
      <description>Inilabas at in-open source ng Xiaomi noong Disyembre 16 ang MiMo-V2-Flash, isang MoE base language model na may 309B total parameters at 15B activated parameters; ang weights ay in-upload sa Hugging Face sa ilalim ng MIT license.</description>
      <category>Paglabas ng modelo</category>
      <source url="https://mimo.xiaomi.com/mimo-v2-flash/index.html">小米 MiMo：官网发布与博客</source>
    </item>
    <item>
      <title>Nagpasalamat ang Tencent Hunyuan sa Artificial Analysis sa evaluation ng Hy4, nakakuha ang modelo ng 35 puntos sa Intelligence Index v5</title>
      <link>https://new.nat6.net/fil/news/fj4zi12biz0b62ckhzg5awcxx/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/fj4zi12biz0b62ckhzg5awcxx/</guid>
      <pubDate>Thu, 05 Nov 2026 10:23:42 +0000</pubDate>
      <description>Ibinahagi at pinasalamatan ng Tencent Hunyuan ang Artificial Analysis sa evaluation ng kanilang pinakabagong model na Hy4. Nakakuha ang Hy4 ng 35 puntos sa Artificial Analysis Intelligence Index v5, pangalawa lamang sa mga modelo ng Chinese lab sa likod ng Kimi K3.1 Pro (46 puntos), kapantay ng Gro…</description>
      <category>Paglabas ng modelo</category>
      <source url="https://x.com/TencentHunyuan/status/2118288137973854397">X：腾讯混元 (@TencentHunyuan)</source>
    </item>
    <item>
      <title>Inilabas ng OpenAI ang GPT-6 Cyber na modelo para sa cybersecurity</title>
      <link>https://new.nat6.net/fil/news/u904yv6dn7sw6lwjgxxedqd3b/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/u904yv6dn7sw6lwjgxxedqd3b/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <description>Inilabas ng OpenAI ang modelong pang-cybersecurity na GPT-6 Cyber, na sinanay para sa pagtuklas, pagpapatunay, at pag-aayos ng kahinaan, at binuksan sa mga security team na naaprubahan sa pamamagitan ng Daybreak Red/Blue platform.</description>
      <category>Paglabas ng modelo</category>
      <category>OpenAI / ChatGPT</category>
      <category>Paglabas ng mga modelo</category>
      <category>Kaligtasan at alignment</category>
      <source url="https://openai.com/index/introducing-gpt-6-cyber">OpenAI：官网动态（RSS · 排除企业/客户案例）</source>
    </item>
    <item>
      <title>Inilabas ng Thinking Machines Lab ang Inkling 2 Small: 594B parameter MoE model, Apache 2.0 weights sa susunod na linggo</title>
      <link>https://new.nat6.net/fil/news/bu8612ca3e1uyxjppk56ef1wl/</link>
      <guid isPermaLink="true">https://new.nat6.net/fil/news/bu8612ca3e1uyxjppk56ef1wl/</guid>
      <pubDate>Thu, 05 Nov 2026 07:00:00 +0000</pubDate>
      <description>Inilabas ng Thinking Machines Lab ang ikalawang henerasyong modelong Inkling 2 Small, isang MoE model na may 594B total parameters at 17B active parameters, sumusuporta sa image input at variable effort, at may score na 31 sa Artificial Analysis Intelligence Index.</description>
      <category>Paglabas ng modelo</category>
      <category>Agent AI agent</category>
      <category>Paglabas ng mga modelo</category>
      <category>AI coding</category>
      <category>Multimodal</category>
      <source url="https://thinkingmachines.ai/news/inkling-2-small/">Thinking Machines Lab：News（网页）</source>
    </item>
  </channel>
</rss>
