<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Free Router LLM News</title>
    <link>https://free-router.com/en/news/</link>
    <description>A weekly read on the LLM models, tooling, and ecosystem moves worth knowing — and what they mean if you run your own gateway.</description>
    <language>en-US</language>
    <atom:link href="https://free-router.com/en/news/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>A model guide for the GPT-6 family</title>
      <link>https://free-router.com/en/news/model-guide-gpt6/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/model-guide-gpt6/</guid>
      <description>OpenAI published a selection guide for the GPT-6 family and a DevDay recap in the same week, alongside a lower-cost model aimed at coding and computer interaction. A catalogue that keeps multiplying is a problem your gateway has to absorb.</description>
      <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>
      <source url="https://openai.com/news/">OpenAI</source>
      <category>Model selection</category><category>OpenAI</category>
    </item>
    <item>
      <title>Assistants that keep working: from one request to sustained execution</title>
      <link>https://free-router.com/en/news/proactive-assistants/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/proactive-assistants/</guid>
      <description>OpenAI introduced dots, an assistant that continues complex tasks on its own, while Anthropic and OpenAI enterprise stories point the same way. Request/response is giving way to agents that carry on without you.</description>
      <pubDate>Thu, 01 Oct 2026 00:00:00 GMT</pubDate>
      <source url="https://openai.com/news/">OpenAI</source>
      <category>Agents</category><category>Product trends</category>
    </item>
    <item>
      <title>Distillation under attack: OpenAI discloses a coordinated extraction effort</title>
      <link>https://free-router.com/en/news/model-distillation-defense/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/model-distillation-defense/</guid>
      <description>OpenAI published details of an organised model-distillation campaign and the anti-distillation work that followed it. For local-first users the interesting part is not the cat-and-mouse, but where your prompts travel.</description>
      <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
      <source url="https://openai.com/news/">OpenAI</source>
      <category>Security</category><category>Model ecosystem</category>
    </item>
    <item>
      <title>Research watch: constant-size scene memory and temporal reasoning in video models</title>
      <link>https://free-router.com/en/news/video-world-models/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/video-world-models/</guid>
      <description>Three papers on Hugging Face in two days — fixed six-plane low-rank memory for long-horizon video generation, re-injecting fading temporal activations at inference, and jointly training a latent tokenizer with a flow-based dynamics model. All three chase the same bottleneck.</description>
      <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
      <source url="https://huggingface.co/papers">Hugging Face Papers</source>
      <category>Research</category><category>Multimodal</category>
    </item>
    <item>
      <title>Open weights and sovereign AI: Mistral raises €3B</title>
      <link>https://free-router.com/en/news/open-weight-sovereign-ai/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/open-weight-sovereign-ai/</guid>
      <description>Mistral closed a €3B Series D at a valuation above €21B and, in the same stretch, opened a Munich hub and announced work on European in-region inference. Open-weight models are moving from alternative to default.</description>
      <pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate>
      <source url="https://mistral.ai/news">Mistral AI</source>
      <category>Open models</category><category>Model ecosystem</category>
    </item>
    <item>
      <title>Evaluation is moving inside the system</title>
      <link>https://free-router.com/en/news/evaluation-and-safeguards/</link>
      <guid isPermaLink="true">https://free-router.com/en/news/evaluation-and-safeguards/</guid>
      <description>Anthropic and Accenture are collaborating on evaluation embedded into AI systems, alongside a life-sciences verification programme and enterprise frontier safeguards. The shared signal: whether a model works is settled on deployment, not on a leaderboard.</description>
      <pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate>
      <source url="https://www.anthropic.com/news">Anthropic</source>
      <category>Evaluation</category><category>Enterprise</category>
    </item>
  </channel>
</rss>