routerhubSI/uncensored-vs-abliterated-ai

选用指南

AI 根据 README 整理 · 2026/10/30 02:16:16

适用场景

适用于需要选择uncensored或abliterated模型、比较价格和拒绝行为的场景，如研究、创意或代理工作流。

查看 README 依据

| **Protocol** | Chat Completions | Chat Completions | Qwen 3.8 Max Uncensored is **25% more expensive** than the standard entry on all three rates. There is no Qwen abliterated entry in the current catalogue, so the choice is standard vs uncensored. --- ## When to use abliterated - You need a **weight-level** intervention, not just a prompt-level persona. - Standard prompts still hit refusals in legitimate research, creative or agent workflows. - You can accept possible changes in tone, instruction following or tool-call behavior.

使用方式

通过RouterHub API使用，需设置API密钥和基础URL，使用OpenAI客户端调用Chat Completions协议。

查看 README 依据

| GLM 5.3 Flash Uncensored | `glm-5.3-flash-uncensored` | $0.15 | $0.50 | $0.013453 | | Qwen 3.8 Max | `qwen3.8-max` | $0.20 | $0.60 | $0.025 | | Qwen 3.8 Max Uncensored | `qwen3.8-max-uncensored` | $0.25 | $0.75 | $0.03125 | Published RouterHub rates, checked 2026-10-27. Prices can change. --- ## Quickstart ```bash export ROUTERHUB_API_KEY="sk-..." export ROUTERHUB_BASE_URL="https://.../v1" ``` ```python import os from openai import OpenAI client = OpenAI( api_key=os.environ["ROUTERHUB_API_KEY"], base_url=os.environ["ROUTERHUB_BASE_URL"], )

response = client.chat.completions.create( model="MODEL_ID_FROM_DASHBOARD", messages=[ {"role": "system", "content": "You are a precise research assistant."}, {"role": "user", "content": "Compare these two model choices for my workload: ..."}, ], ) print(response.choices[0].message.content) ``` The DeepSeek, GLM and Qwen entries above use **Chat Completions**. Confirm the supported protocol for the exact model ID in the [catalogue](https://routerhub.si/models). Full setup: [API quickstart](./docs/09-api-quickstart.md). --- ## How to evaluate them

部署要求

需要RouterHub API密钥，并设置环境变量ROUTERHUB_API_KEY和ROUTERHUB_BASE_URL。

查看 README 依据

| GLM 5.3 Flash Uncensored | `glm-5.3-flash-uncensored` | $0.15 | $0.50 | $0.013453 | | Qwen 3.8 Max | `qwen3.8-max` | $0.20 | $0.60 | $0.025 | | Qwen 3.8 Max Uncensored | `qwen3.8-max-uncensored` | $0.25 | $0.75 | $0.03125 | Published RouterHub rates, checked 2026-10-27. Prices can change. --- ## Quickstart ```bash export ROUTERHUB_API_KEY="sk-..." export ROUTERHUB_BASE_URL="https://.../v1" ``` ```python import os from openai import OpenAI client = OpenAI( api_key=os.environ["ROUTERHUB_API_KEY"], base_url=os.environ["ROUTERHUB_BASE_URL"], )

已知限制

abliterated不保证零拒绝，uncensored不保证无限制输出，行为因模型ID、提示和提供商政策而异。

查看 README 依据

In the current RouterHub catalogue, the **uncensored** entries are often cheaper than the **abliterated** entries for the same model family. If prompt-level control is enough for your use case, uncensored can be the better value. If you need a weight-level intervention, choose abliterated — and test it. > **Important:** “abliterated” is not a guarantee of zero refusals, and “uncensored” is not a guarantee of unrestricted output. Behaviour depends on the exact model ID, prompt and provider policy. --- ## Uncensored vs abliterated — the difference | | Abliterated | Uncensored |