
Refuseless: Open and Abliterated LLMs
Very powerful model competing with frontier at much lower cost.
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Very powerful model competing with frontier at much lower cost.
Scores for the base checkpoint, Z AI's GLM-5.3-Flash, not a Refuseless eval. Source: Artificial Analysis
Setup takes about a minute. Pick your agent:
OpenCodePi
1Add Refuseless to your OpenCode config
Mac / Linux~/.config/opencode/opencode.json
Windows%USERPROFILE%.config\opencode\opencode.json
{ "$schema": "https://opencode.ai/config.json", "model": "refuseless/glm-5.3-abliterated", "provider": { "refuseless": { "npm": "@ai-sdk/openai-compatible", "name": "Refuseless", "options": { "baseURL": "https://api.refuseless.com/v1" }, "models": { "glm-5.3-abliterated": { "name": "GLM 5.3 Abliterated", "tools": true, "reasoning": true, "limit": { "context": 1048576, "output": 131072 } }, "glm-5.3-flash-abliterated": { "name": "GLM 5.3 Flash Abliterated", "tools": true, "reasoning": true, "limit": { "context": 1048576, "output": 131072 } }, "qwen3.8-flash-next-abliterated": { "name": "Qwen3.8 Flash Next Abliterated", "tools": true, "reasoning": true, "limit": { "context": 262144, "output": 65536 } }, "deepseek-v4.1-flash-abliterated": { "name": "DeepSeek V4.1 Flash Abliterated", "tools": true, "reasoning": true, "limit": { "context": 1048576, "output": 262144 } }, "deepseek-v4.1-flash-fast": { "name": "DeepSeek V4.1 Flash 1017 (Safeguards)", "tools": true, "reasoning": true, "limit": { "context": 1048576, "output": 131072 } } } } } }
Copy whole fileCopiedFailedCopy whole fileCopy highlighted blockCopiedFailedCopy highlighted block
New file: copy the whole file. Already have one: copy only the highlighted block and paste it inside the existing "provider" object, keeping your other providers.
Start opencode, type /connect, choose Refuseless and paste your key. If Refuseless isn't listed, pick Other and enter refuseless as the provider id.
Use the key from your API keys page. It starts with abl_.
OpenCode saves this key separately, and it takes priority over any apiKey in the config. If you create a new key later, run /connect again, or OpenCode keeps sending the old one and you get Invalid API key.
opencode --model refuseless/glm-5.3-abliterated
CopyCopiedFailedCopy
Or use /models in OpenCode and pick Refuseless › GLM 5.3 Abliterated. Send Say ok. to check it works. All 5 lineup models are in the config, so you can switch any time.
Same request shape as OpenAI chat completions. Point the base URL at Refuseless, set a lineup id, send the key.
curl https://api.refuseless.com/v1/chat/completions
-H "Authorization: Bearer $REFUSELESS_API_KEY"
-H "Content-Type: application/json"
-d '{
"model": "glm-5.3-abliterated",
"messages": [{"role": "user", "content": "Write a tight system prompt."}]
}'
Host and path are the intended contract. Auth is the API key.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。