Qwen 3.8 27B (Fully Tested on Local Mac 48GB) + Hermes: I might CANCEL Claude & Codex NOW!

Thank you to TestSprite for sponsoring this video: https://www.testsprite.com/?via=aicodingki-dw Github repo for TestSprite: https://github.com/TestSprite/testsprite-cli In this video, I'll be telling you about Qwen 3.8 27B, a new open-weight local AI model that is built for agent workflows, vision, computer use, long context, and tool calling. I’ll also explain why I did not run it on KingBench, how it compares to frontier models, and how to run it properly with Ollama, LM Studio, and Hermes Agent. -- Key Takeaways: 🚀 Qwen 3.8 27B is an Apache 2.0 open-weight model that can run locally on your own hardware. 🧠 It is not meant to compete as a frontier reasoning model, but it performs extremely well as a local agent brain. 👁️ The model is a native vision-language model with support for images, documents, diagrams, screens, and video. 📏 Qwen 3.8 27B supports a massive 262K native context window, extendable up to 1 million tokens with RoPE scaling. 💻 It performs especially well on agent, coding, browser, computer-use, and multimodal software engineering benchmarks. ⚙️ Ollama and LM Studio both offer simple ways to run the model locally, with day-one quantized model support. 🔧 Setting the context length properly is critical, because the default settings may waste most of the model’s long-context capability. 🧰 Tool calling works smoothly through Ollama and LM Studio, making Qwen 3.8 27B a strong fit for Hermes Agent. 🤖 Hermes Agent can use Qwen 3.8 27B through a custom OpenAI-compatible local endpoint for a complete local agent stack. 👍 Overall, Qwen 3.8 27B is one of the most impressive local agent models available, especially for long, tool-heavy workflows.