PulseAugur
中
实时 17:04:20
English(EN) A 3x Faster Local LLM That Can't Finish the Job — Two Days of MoE vs Dense on Real Tasks

快速MoE LLM Ornith 1.0 35B在真实世界测试中败给密集模型Qwen3.6

作者 PulseAugur 编辑部 · [1 个来源] ·

一个本地的混合专家（MoE）模型Ornith 1.0 35B，最初以其速度和编码能力令人印象深刻，其表现比密集模型Qwen3.6 27B快三倍。然而，这种优势并未持续多久，因为Ornith在通用任务中开始显现出显著的缺陷，包括捏造信息和误解数据。严格的测试表明，Ornith在完成任务方面持续失败，经常留下不完整或无意义的代码，而较慢的密集模型则可靠地完成了相同的任务。 AI

影响 MoE模型可能提供速度优势，但在复杂任务中的可靠性和事实准确性方面可能存在挑战。

排序理由 在真实世界任务中对比两个LLM，并非新模型发布。

在 dev.to — LLM tag 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

快速MoE LLM Ornith 1.0 35B在真实世界测试中败给密集模型Qwen3.6

本文如何被排名

Signal score
44 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
在真实世界任务中对比两个LLM，并非新模型发布。
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
model release, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. dev.to — LLM tag TIER_1 English(EN) · kohei_wada ·

    A 3x Faster Local LLM That Can't Finish the Job — Two Days of MoE vs Dense on Real Tasks

    <h2> Introduction </h2> <p>I run ollama on a home RTX 5090 (32GB) and use it through opencode. My default model<br /> had long been the dense Qwen3.6 27B, but when I tried <strong>Ornith 1.0 35B</strong> (35B/A3B),<br /> a coding-focused MoE model, the numbers were too good to ig…

相关实体

相关话题