Nous Portal 免费模型完整使用指南
结论先行:Nous Research 提供 3 个完全免费的大模型(Step-3.7-Flash、Nemotron-3-Ultra、Owl-Alpha),支持标准 OpenAI 兼容 API,可直接用于 Hermes、OpenClaw、Trae、Continue、Cline、Aider 等所有 AI 工具。
一、什么是 Nous Portal?
Nous Portal 是 AI 研究机构 Nous Research 提供的模型推理平台,聚合了 266+ 开源大模型,提供统一的 OpenAI 兼容 API 接口。
| 项目 | 详情 |
| 官网 | https://portal.nousresearch.com |
| API 地址 | https://inference-api.nousresearch.com/v1 |
| 兼容格式 | OpenAI API 格式 |
| 免费模型 | 3 个(价格 = 0) |
| 认证方式 | OAuth 设备码流程 |
二、申请 Token 教程
方式一:通过 Hermes 自动获取(推荐)
# 第一步:添加 Nous 认证
hermes auth add nous
执行后终端显示:
Please visit: https://portal.nousresearch.com/device
Enter code: XXXX-XXXX
操作步骤:
1. 打开浏览器访问 https://portal.nousresearch.com/device
2. 输入终端显示的设备码(如 ABCD-1234)
3. 点击"Authorize"授权
4. 终端自动完成认证,Token 保存到 ~/.hermes/auth.json
# 第二步:查看认证状态
hermes auth list
# 第三步:切换到 Nous Provider
hermes model
# 选择 nous → 选择 stepfun/step-3.7-flash:free
方式二:Nous Portal 网站手动获取
1. 浏览器打开 https://portal.nousresearch.com
2. 点击右上角 Sign Up 注册账号(支持 Google/GitHub 登录)
3. 登录后进入 Dashboard 或 API Keys 页面
4. 点击 Create API Key / Generate Token
5. 复制生成的 Token
Token 有效期
• Access Token:约 15 分钟(短期)
• Refresh Token:长期有效
• 自动刷新:Hermes 会自动使用 refresh_token 刷新
• 其他工具:部分工具不支持自动刷新,需要定期重新获取
三、免费模型完整列表
通过 API 查询,共 3 个免费模型(价格 = 0):
| 模型 | 厂商 | 上下文 | 特点 |
| stepfun/step-3.7-flash:free | 阶跃星辰 | 256K | 速度快,中文好 |
| nvidia/nemotron-3-ultra:free | NVIDIA | 1M | 超长上下文 |
| openrouter/owl-alpha | OpenRouter | 1M | 超长上下文 |
⚠️ 注意:价格 = 0 不等于无限免费,可能有速率限制。:free 后缀是明确标记的免费模型。
step-3.7-flash:free 实测评测
我们对 Nous Portal 提供的免费模型中综合表现最好的 step-3.7-flash:free 进行了全面实测,涵盖速度、推理、代码、中文理解等维度。该模型是一个 Reasoning 模型(先内部推理再输出最终答案),类似 DeepSeek-R1 的推理链机制。
速度基准测试
| 测试项 | 延迟(s) | 输出Tokens | 速度(t/s) | 状态 |
| 简单问答 | 4.70 | 85 | 18.1 | stop |
| 逻辑推理 | 13.13 | 766 | 58.3 | stop |
| 长文本生成 | 25.46 | 2281 | 89.6 | stop |
| 中文理解 | 9.25 | 457 | 49.4 | stop |
| 数学推理 | 12.88 | 871 | 67.6 | stop |
| 代码生成 | — | — | — | 503 不可用 |
| 简单计算(引导) | 2.78 | 34 | 12.2 | stop |
汇总:平均延迟 13.1s | 平均生成速度 56.6 t/s | 速度范围 18-90 t/s
能力测试
| 测试项 | 结果 | 评价 |
| 简单问答 | "Paris" | 正确 |
| 逻辑推理 | 正确解释三段论不必然性 | 推理严谨 |
| 长文本生成 | 量子计算解释完整(1390 chars) | 质量良好 |
| 中文理解 | 表达流畅自然 | 中文优秀 |
| 数学推理 | 方程求解 ball=$0.05,步骤清晰 | 数学正确 |
| 代码生成 | 503 错误 | 暂时不可用 |
模型特征
| 特征 | 说明 |
| 模型类型 | Reasoning 模型,输出分离为 reasoning + content |
| reasoning/content 比 | 约 5:1(大量 token 用于内部思考) |
| 中文支持 | 良好,表达自然 |
| 数学/逻辑 | 正确率高,推理过程透明 |
| 稳定性 | 代码生成 503,可能有并发/路由限制 |
| 免费额度 | :free 标签,完全免费 |
综合评价
定位:轻量级免费 reasoning 模型,适合需要透明推理过程的场景。约处于 Qwen2.5-7B 到 DeepSeek-V3 之间,免费模型中推理能力较好。
优势:中文流畅、逻辑推理严谨、数学准确、完全免费零门槛
短板:延迟偏高(reasoning 开销大)、速度波动明显(18~90 t/s)、代码生成接口不稳定
建议:日常对话、中文问答、逻辑推理场景首选;高强度代码生成建议搭配其他模型
四、在各 AI 工具中配置
核心配置三要素
| 项目 | 值 |
| API Base URL | https://inference-api.nousresearch.com/v1 |
| API Key | Nous Portal 的 OAuth token |
| 模型名 | stepfun/step-3.7-flash:free |
4.1 Hermes Agent
# 方法一:交互式向导
hermes model
# 方法二:手动认证
hermes auth add nous
# 方法三:一行命令
hermes chat -m stepfun/step-3.7-flash:free --provider nous -q "你好"
4.2 OpenClaw
# ~/.openclaw/config.yaml
provider: openai-compatible
base_url: https://inference-api.nousresearch.com/v1
api_key: your_nous_token_here
model: stepfun/step-3.7-flash:free
4.3 Trae(字节跳动 AI IDE)
方式一:设置界面
1. 打开 Trae → 设置 → AI 模型
2. 选择"自定义模型"或"OpenAI 兼容"
3. 填写 API 地址、API Key、模型名
方式二:配置文件
// ~/.trae/config.json
{
"llm": {
"provider": "openai",
"baseUrl": "https://inference-api.nousresearch.com/v1",
"apiKey": "your_nous_token_here",
"model": "stepfun/step-3.7-flash:free"
}
}
4.4 Continue(VS Code / JetBrains 插件)
// ~/.continue/config.json
{
"models": [
{
"title": "Nous Portal - Step 3.7 Flash",
"provider": "openai",
"model": "stepfun/step-3.7-flash:free",
"apiBase": "https://inference-api.nousresearch.com/v1",
"apiKey": "your_nous_token_here"
}
]
}
4.5 Cline(VS Code 插件)
1. VS Code → Cline 插件 → 设置
2. API Provider 选择 "OpenAI Compatible"
3. 填写 Base URL、API Key、Model ID
4.6 Aider(终端 AI 编程工具)
# 环境变量方式
export OPENAI_API_BASE=https://inference-api.nousresearch.com/v1
export OPENAI_API_KEY=your_token
# 启动 Aider
aider --model stepfun/step-3.7-flash:free
4.7 Cherry Studio / ChatBox
| 设置项 | 值 |
| API 类型 | OpenAI Compatible |
| API 地址 | https://inference-api.nousresearch.com/v1 |
| 模型 | stepfun/step-3.7-flash:free |
4.8 Python / Node.js 直接调用
Python
from openai import OpenAI
client = OpenAI(
base_url="https://inference-api.nousresearch.com/v1",
api_key="your_nous_token_here"
)
response = client.chat.completions.create(
model="stepfun/step-3.7-flash:free",
messages=[{"role": "user", "content": "你好"}]
)
print(response.choices[0].message.content)
Node.js
import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://inference-api.nousresearch.com/v1',
apiKey: 'your_nous_token_here'
});
const response = await client.chat.completions.create({
model: 'stepfun/step-3.7-flash:free',
messages: [{ role: 'user', content: 'Hello!' }]
});
console.log(response.choices[0].message.content);
五、故障排查
问题 1:认证失败
# 故障排查:认证失败
hermes auth remove nous
hermes auth add nous
问题 2:请求超时
# 检查网络
curl -I https://inference-api.nousresearch.com/v1/models
# 国内可能需要代理
export https_proxy=http://your-proxy:port
问题 3:429 限流
• 等待几分钟后重试
• 切换到其他免费模型
• 高峰期避免高并发
六、方案对比
| 方案 | 免费模型 | 质量 | 难度 |
| Nous Portal | 3 个 | 中等 | 简单 |
| OpenRouter Free | 10+ 个 | 较高 | 中等 |
| Google Gemini | 2-3 个 | 较高 | 中等 |
| DeepSeek 官方 | 0(付费) | 顶级 | 简单 |
七、快速参考
# 申请 Token
hermes auth add nous
# 配置
hermes model
hermes auth list
# 使用
hermes chat -q "你的问题"
hermes chat -m stepfun/step-3.7-flash:free --provider nous
# 调试
hermes doctor
最后更新:2026-06-08 | Hermes Agent 自动生成
Complete Usage Guide for the Nous Portal Free Models
Conclusion first: Nous Research provides 3 completely free large models (Step-3.7-Flash, Nemotron-3-Ultra, Owl-Alpha), supporting a standard OpenAI-compatible API that can be used directly with Hermes, OpenClaw, Trae, Continue
1. What is Nous Portal?
Nous Portal is a model inference platform provided by the AI research institute Nous Research, aggregating 266+ open-source large models and offering a unified OpenAI-compatible API interface.
2. Tutorial for applying for a Token
Method 1: obtain it automatically through Hermes (recommended)
# Step 1: add Nous authentication
hermes auth add nous
After running, the terminal displays:
Please visit: https://portal.nousresearch.com/device
Enter code: XXXX-XXXX
Steps:
1. Open a browser and go to https://portal.nousresearch.com/device
2. Enter the device code shown in the terminal (such as ABCD-1234)
3. Click 'Authorize' to authorise
4. The terminal completes authentication automatically, and the Token is saved to ~/.hermes/auth.json
# Step 2: check authentication status
hermes auth list
# Step 3: switch to the Nous Provider
hermes model
# Choose nous → choose stepfun/step-3.7-flash:free
Method 2: obtain it manually from the Nous Portal website
1. Open https://portal.nousresearch.com in a browser
2. Click Sign Up in the top right to register (Google/GitHub login supported)
3. After logging in, go to the Dashboard or API Keys page
4. Click Create API Key / Generate Token
5. Copy the generated Token
Token validity
• Access Token: about 15 minutes (short-term)
• Refresh Token: long-term valid
• Auto refresh: Hermes automatically uses refresh_token to refresh
• Other tools: some tools do not support auto refresh and need periodic re-obtaining
3. Complete list of free models
Querying via the API, there are 3 free models in total (price = 0):
⚠️ Note: price = 0 does not mean unlimited free usage; there may be rate limits. The :free suffix explicitly marks a free model.
Hands-on evaluation of step-3.7-flash:free
We ran a comprehensive hands-on evaluation of step-3.7-flash:free, the best all-round performer among the free models from Nous Portal, covering speed, reasoning, code and Chinese comprehension. This model is a Reasoning model (reasoning internally first, then outputting the final answer), similar to DeepSeek-R1.
Speed benchmark
Summary: average latency 13.1s | average generation speed 56.6 t/s | speed range 18-90 t/s
Capability testing
Model characteristics
Overall assessment
Positioning: a lightweight free reasoning model, suited to scenarios needing a transparent reasoning process. Roughly between Qwen2.5-7B and DeepSeek-V3, with good reasoning ability among free models.
Strengths: fluent Chinese, rigorous logical reasoning, accurate mathematics, completely free with zero barrier
Weaknesses: latency runs high (reasoning overhead), speed fluctuates noticeably (18-90 t/s), code generation interface is unstable
Recommendation: first choice for everyday conversation, Chinese Q&A and logical reasoning; for high-intensity code generation, pair it with another model
4. Configuration in each AI tool
The three core configuration elements
4.1 Hermes Agent
# Method 1: interactive wizard
hermes model
# Method 2: manual authentication
hermes auth add nous
# Method 3: one-line command
hermes chat -m stepfun/step-3.7-flash:free --provider nous -q "Hello"
4.2 OpenClaw
# ~/.openclaw/config.yaml
provider: openai-compatible
base_url: https://inference-api.nousresearch.com/v1
api_key: your_nous_token_here
model: stepfun/step-3.7-flash:free
4.3 Trae (ByteDance AI IDE)
Method 1: settings interface
1. Open Trae → Settings → AI Models
2. Choose 'Custom Model' or 'OpenAI Compatible'
3. Fill in the API address, API Key and model name
Method 2: config file
// ~/.trae/config.json
{
"llm": {
"provider": "openai",
"baseUrl": "https://inference-api.nousresearch.com/v1",
"apiKey": "your_nous_token_here",
"model": "stepfun/step-3.7-flash:free"
}
}
4.4 Continue (VS Code / JetBrains plugin)
// ~/.continue/config.json
{
"models": [
{
"title": "Nous Portal - Step 3.7 Flash",
"provider": "openai",
"model": "stepfun/step-3.7-flash:free",
"apiBase": "https://inference-api.nousresearch.com/v1",
"apiKey": "your_nous_token_here"
}
]
}
4.5 Cline (VS Code plugin)
1. VS Code → Cline plugin → Settings
2. Set API Provider to "OpenAI Compatible"
3. Fill in Base URL, API Key and Model ID
4.6 Aider (terminal AI coding tool)
# Environment variable method
export OPENAI_API_BASE=https://inference-api.nousresearch.com/v1
export OPENAI_API_KEY=your_token
# Launch Aider
aider --model stepfun/step-3.7-flash:free
4.7 Cherry Studio / ChatBox
4.8 Direct calls from Python / Node.js
Python
from openai import OpenAI
client = OpenAI(
base_url="https://inference-api.nousresearch.com/v1",
api_key="your_nous_token_here"
)
response = client.chat.completions.create(
model="stepfun/step-3.7-flash:free",
messages=[{"role": "user", "content": "Hello"}]
)
print(response.choices[0].message.content)
Node.js
import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://inference-api.nousresearch.com/v1',
apiKey: 'your_nous_token_here'
});
const response = await client.chat.completions.create({
model: 'stepfun/step-3.7-flash:free',
messages: [{ role: 'user', content: 'Hello!' }]
});
console.log(response.choices[0].message.content);
5. Troubleshooting
Problem 1: authentication failure
# Troubleshooting: authentication failure
hermes auth remove nous
hermes auth add nous
Problem 2: request timeout
# Check the network
curl -I https://inference-api.nousresearch.com/v1/models
# A proxy may be needed in China
export https_proxy=http://your-proxy:port
Problem 3: 429 rate limit
• Wait a few minutes and retry
• Switch to another free model
• Avoid high concurrency at peak times
6. Scheme comparison
7. Quick reference
# Apply for a Token
hermes auth add nous
# Configure
hermes model
hermes auth list
# Use
hermes chat -q "Your question"
hermes chat -m stepfun/step-3.7-flash:free --provider nous
# Debug
hermes doctor
Last updated: 2026-06-08 | Auto-generated by Hermes Agent