<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Hoja — Blog</title><description>Engineering write-ups from Hoja: open-source tools, model choices, and the decisions behind our projects.</description><link>https://hoja-solutions.github.io/</link><item><title>Qwen3.8-27B: the same architecture, fourteen points better</title><link>https://hoja-solutions.github.io/blog/qwen38-27b-local/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/qwen38-27b-local/</guid><description>Alibaba shipped a 27B model in August that scores 52 on the Artificial Analysis index against its predecessor&apos;s 38, with an identical network. What that means for running a capable model on one graphics card, and what it costs in memory and tokens.</description><pubDate>Mon, 24 Aug 2026 12:00:00 GMT</pubDate><category>AI</category><category>Open Source</category><category>Hardware</category></item><item><title>The frontier after Kimi K3: four points apart, three times the price</title><link>https://hoja-solutions.github.io/blog/frontier-after-kimi-k3/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/frontier-after-kimi-k3/</guid><description>The top four models on the Artificial Analysis index sit within four points of each other and vary threefold in cost per task. The cheapest of them is the one you can download. What changed in July was pricing power, not capability.</description><pubDate>Wed, 05 Aug 2026 12:00:00 GMT</pubDate><category>AI</category><category>Open Source</category><category>Hardware</category></item><item><title>The open-weights letter: what 25 companies asked Washington for, and what they left out</title><link>https://hoja-solutions.github.io/blog/open-weights-letter/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/open-weights-letter/</guid><description>NVIDIA, Microsoft, Meta, and 22 others signed a letter arguing the US should not restrict downloadable models. We read the three pages, checked the one claim we can verify from our own work, and noted what the letter avoids.</description><pubDate>Fri, 24 Jul 2026 12:00:00 GMT</pubDate><category>AI</category><category>Open Source</category><category>Policy</category></item><item><title>From a Raspberry Pi to a DGX Spark: the state of local models in 2026</title><link>https://hoja-solutions.github.io/blog/state-of-local-models/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/state-of-local-models/</guid><description>The hardware to hold a capable model got cheap, and the models got small enough to fit. What runs on the machines in your bedroom, and the two numbers that decide it.</description><pubDate>Thu, 23 Jul 2026 00:00:00 GMT</pubDate><category>AI</category><category>Open Source</category><category>Hardware</category></item><item><title>agent-stdlib: an installable agent toolkit</title><link>https://hoja-solutions.github.io/blog/agent-stdlib-installable-toolkit/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/agent-stdlib-installable-toolkit/</guid><description>Anthropic&apos;s engineering blog describes how to build agents in production. agent-stdlib packages those procedures into installable skills, MCP servers, and safety hooks you can run in any harness.</description><pubDate>Wed, 22 Jul 2026 12:00:00 GMT</pubDate><category>Engineering</category><category>Open Source</category><category>AI Agents</category></item><item><title>Why we build open-source</title><link>https://hoja-solutions.github.io/blog/why-we-build-open-source/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/why-we-build-open-source/</guid><description>This website is a public GitHub repository, and so is the rest of our stack. The practical case for building on open source.</description><pubDate>Wed, 22 Jul 2026 12:00:00 GMT</pubDate><category>Open Source</category><category>Engineering</category></item><item><title>Building Hoja&apos;s website with an open-source LLM</title><link>https://hoja-solutions.github.io/blog/open-source-llm-website/</link><guid isPermaLink="true">https://hoja-solutions.github.io/blog/open-source-llm-website/</guid><description>We built our production website with Qwen3.6-27B, an open-source model from Alibaba. What frontier open-weight models can do now, and where you still need a person in the loop.</description><pubDate>Tue, 21 Jul 2026 12:00:00 GMT</pubDate><category>AI</category><category>Engineering</category><category>Open Source</category></item></channel></rss>