体验记录
Qwen
| 模型 | 标签 | 评价 |
| Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distill-heretic-v3.i1-Q4_K_M.gguf | English heretic abliteration uncensored imatrix conversational | opus 4B 体验还不错，照样能打 |
| qwen3.5-4b-python-coder-q4_k_m.gguf | 虽然是coder模型，continue补全返回总是空 | |
| Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated-heretic.i1-Q4_K_S.gguf | English abliterated uncensored Claude reasoning chain-of-thought Dense heretic decensored imatrix conversational | chat没有任何问题 opencode+omo: Cannot determine type of 'item' |
| Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING.i1-Q5_K_M.gguf | English Chinese fine tune creative creative writing fiction writing plot generation sub-plot generation story generation scene continue storytelling fiction story science fiction romance all genres story writing vivid prosing vivid writing fiction roleplaying bfloat16 all use cases unsloth heretic uncensored abliterated imatrix conversational | opencode+omo: Cannot determine type of 'item' |
| Qwen3.5-9B-Claude-4.6-Opus-Deckard-V4.2-Uncensored-Heretic-Thinking.i1-Q4_K_M.gguf | ||
| Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated-Q4_K_M.gguf | ||
| Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-heretic.i1-Q6_K.gguf | 功能交互正常，很能打 | |
| Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GGUF | ||
| Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated.i1-Q4_K_S.gguf | 有丢失逻辑的感觉 | |
| Qwen3.5-35B-A3B-heretic-Opus-4.6-Distilled.i1-Q5_K_M.gguf | 存在prefill is incompatible问题，llama.cpp最新版+jinja | |
| Qwen3.5-35B-A3B-heretic-v2-Opus-4.6-Distilled.i1-Q5_K_M.gguf | 存在prefill is incompatible问题，llama.cpp最新版+jinja | |
| Qwen3.6-35B-A3B-Abliterated-Heretic-Q4_K_M.gguf | 存在prefill is incompatible问题，llama.cpp最新版+jinja 有时候编程任务计算返回空 | |
| Qwen3.6-35B-A3B-Opus-Q4_K_S.gguf | ||
| rico03/Qwen3.6-35B-Opus-Reasoning-GGU | ||
| hesamation/Qwen3.6-35B-A3B-Claude-4.6-Opus-Reasoning-Distilled.Q4_K_M.gguf | ||
| XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic.i1-Q4_K_M.gguf |
Gemma
| 模型 | 标签 | 评价 |
| gemma-4-E4b-it.Q4_K_M.gguf | opencode有unused24问题，训练2025年1月 opencode+omo: Cannot determine type of 'item' | |
| gemma-4-E2B-abliterated-btx-Q4_K_M.gguf | 感觉很一般 | |
| SuperGemma4-31b-abliterated.Q4_K_M.gguf | ||
| gemma-4-31B-it-Claude-Opus-Distill.q4_k_m.gguf’ | 比较严重的 Assistant response prefill is incompatible with enable_thinking. 显卡干崩2次，不如Qwen系列好资源可控 | |
Llama
| 模型 | 标签 | 评价 |
| llama4-dolphin-8b.q4_k_m.gguf | 没啥感受 | |
| Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning.i1-Q4_K_S | 10 languages thinking reasoning instruct heretic uncensored abliterated Claude4.5-Opus creative creative writing fiction writing plot generation sub-plot generation story generation scene continue storytelling fiction story science fiction romance all genres story writing vivid prosing vivid writing fiction roleplaying bfloat16 role play 128k context llama3.3 llama-3 llama-3.3 unsloth finetune imatrix conversational | opencode+omo: Cannot determine type of 'item' |
Nemotron
| 模型 | 标签 | 评价 |
| Nemotron-9B-OpenCode.i1-Q4_K_M.gguf | English Chinese code instruction-tuned software-engineering agent opencode qwen python imatrix conversational | opencode+omo: Cannot determine type of 'item' |
其它
| 模型 | 标签 | 评价 |
| nomic-embed-text-v1.5-Q8_0.gguf | 没啥存在感 |
模型仓库
抱脸 https://huggingface.co/ 美国AI初创公司
魔塔社区 https://modelscope.cn/models 阿里达摩院
鹏城实验室 https://openi.pcl.ac.cn/modelbase/list 深圳科技部
飞桨 https://aistudio.baidu.com/modelsoverview 百度
硅基流动 https://siliconflow.cn/models OneFlow袁进辉
卡格尔 https://www.kaggle.com/models Google开源模型
Grok Grok-2.5（314B MoE） xAI（马斯克） 打破闭源实时能力垄断，倒逼 Meta/Google 加速开源
Llama Llama 4-405B（多模态） Meta 原生多模态，开源生态最广，开发者首选基座
Mistral Mistral Small 4（22B） Mistral AI 推理速度碾压同级，边缘部署首选，重塑轻量开源标准
Gemma Gemma 4（E2B MoE） Google E2B仅1.5GB可手机离线跑，24小时下载破4 亿
Phi Phi-4（14B/38B） Microsoft 小参数强推理，14B版数学/代码超Llama3-70B
DBRX DBRX-132B（MoE） Databricks 首个开源MoE基座，代码/SQL能力登顶，推动开源MoE普及
Nemotron Nemotron-4 340B NVIDIA NVIDIA官方开源旗舰，绑定CUDA生态，企业级部署首选
OLMo OLMo-2-7B/13B Allen 无商用限制，科研界标杆，推动 AI 透明化研究
Qwen Qwen3.5-397B（MoE） 阿里通义 LM Arena盲测国产第一、全球第五，中文能力碾压
DeepSeek DeepSeek-V3.2 深度求索 全面用海光/寒武纪国产芯片训练，脱离CUDA，成本降 60%
GLM GLM-5.1（754B） 智谱 AI SWE-bench Pro 国产第一、全球第三
InternLM InternLM3-102B 上海商汤/浪潮 开源工具调用增强版，国产芯片友好，书生・浦语生态核心