Low-cost AI models anyone can afford

Not enough tokens? Use Newcclm.
A tenth of the official price — as low as 10%.

Flash-grade speed, flagship-grade work. $5 a month — about 17¢ a day.

⚡ Subscribe and get your Newcclm key

$5 / month Card / USDT $5 · Alipay / WeChat ¥30

Billed monthly, delivered instantly after payment. Default newcclm (quality first), newcclm-fast for everyday chat, newcclm-coding for coding agents, newcclm-free for the free tier.

50 left this round · delivered automatically the moment payment succeeds. See all pricing

Supported agents

The agents and clients you already use plug straight in. No switching tools, nothing new to learn.

⚡ newcclm ⚡ DSH ⚡ Cursor ⚡ Cherry Studio ⚡ Chatbox ⚡ NextChat ⚡ Pi Harness ⚡ magpie ⚡ Codex ⚡ OpenCode ⚡ Open Minis ⚡ Oh My Pi ⚡ WorkBuddy ⚡ Anything else
OpenAI Anthropic Google Gemini xAI Mistral Meta Qwen MiniMax Moonshot Perplexity ByteDance Ollama OpenRouter Hugging Face

Models from these vendors all run on our routes. If a tool takes a Base URL and a model name, it works — including Obsidian, Open WebUI and Immersive Translate, anything that accepts a custom OpenAI endpoint. See the general guide.

What does Newcclm solve?

We're trying to put large models in front of as many people as possible. Newcclm is a low-cost subscription at $5 a month — a generous allowance and reliable access to mainstream frontier models, without first having to work out endpoints, parameters and billing. ¥30 a month works too, and small teams can share one plan.

Two tiers, picked by the job

Different work needs different ability. You pick "fast" or "smart" — we handle the rest.

/fast
Fast

Email replies, copy edits, tidying up notes, quick questions. Speed matters more than depth.

/smart
Smart

Complex analysis, long documents, writing code. Slower, but it thinks further.

Why only two? Because most people don't want to choose between a dozen model names — that's a developer's hobby, not a normal person's need. We folded it into two: type /fast or /smart and switch any time.

Works with the tools you already use

Most of the software on your machine can connect directly. No switching tools, nothing new to learn.

Newcclm clientDesktop / terminal Cherry StudioDesktop GUI ChatboxDesktop / mobile NextChatBrowser CursorCode editor DeepSeek HarnessDesktop / terminal Anything elseGeneral compatibility
Whichever tier you name is the tier that runs. We never swap your "smart" for "fast" behind your back, and never downgrade the tier. For the exact fields per tool, the help centre walks through each one.

Which models you can call

Use one of these four names — whichever you send is what runs. Anything else (gpt-4o, claude-* and so on) is rejected.

newcclm
Everyday default

Not sure which to pick? This one. Writing, questions, analysis, tidying up — quality first.

GLM-5.3-Flash · 1M context
newcclm-fast
Faster

For things that need speed more than deep thought: email replies, copy edits, batches.

GLM-5.3-Flash · 1M context
newcclm-coding
Coding / agents

Write code, read long files, run agent loops. Tuned for long files and tool calls.

Qwen3.8-Flash · 1M context
newcclm-free
Free tier

A zero-cost model for trying things out and for jobs that aren't in a hurry.

Free models · limited allowance
There's also a backup route. DeepSeek V4-Flash takes over automatically if the main route fails — that's failover, not a downgrade, and you switch nothing.
Recommended setup

Your main model takes the hard part, we take the subtasks

Planning, decisions and difficult reasoning go to the strongest model you have (say gemini-4-argon). Searching, editing files, running tests and batch tool calls come to newcclm-coding / newcclm-fast. Whichever name a subtask asks for is what the server runs.

Main model

The stronger model owns the goal, the plan and the review.

Subtasks

High-frequency, parallel execution comes to us.

What you name is what runs

Each subagent states newcclm, newcclm-fast or newcclm-coding, and the server never rewrites it.

You can set up exactly this strategy inside magpie — big and small working together.

Setting up common tools

Base URL: https://sccclm.com/v1 Model: newcclm
newcclm-fast for speed · newcclm-coding for code · newcclm is the default · newcclm-free is free
DSH (DeepSeek Harness)Custom provider

In Settings → Models add a custom provider — don't use the built-in DeepSeek / Anthropic entries:

llm-pi-ai:
  providers:
    newcclm:
      api: openai-completions
      baseURL: https://sccclm.com/v1
      apiKeyEnv: NEWCCLM_API_KEY
      models:
        - id: newcclm
Full guide →
CursorOpenAI compatible

Set these in the config (don't pick the Anthropic / DeepSeek presets):

Base URL: https://sccclm.com/v1
Model:    newcclm-coding
API Key:  YOUR_API_KEY
Full guide →
Cherry StudioDesktop app

Add an OpenAI-compatible provider and fill in these three fields:

Provider: OpenAI
Base URL: https://sccclm.com/v1
API Key:  YOUR_API_KEY
Model:    newcclm
Full guide →
Command line / cURLDirect test

To check the endpoint and key on their own, run this:

curl https://sccclm.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{"model":"newcclm","messages":[
      {"role":"user","content":"Hello"}]}'
All guides →

What's underneath

We publish the models openly. Whichever mode you pick, that's the model you get — nothing hidden, nothing swapped.

GEMINI 4 ARGON

前沿推理

100 万 token 上下文 · 文本 / 图像输入
官方 $2.00 输入 / $10.00 输出 每百万 token
已完成适配 newcclm-pro
TENCENT HY4 PREVIEW

编码与 Agent

770B 总参数 / 49B 激活 · 100 万 token 上下文，最长输出 6.4 万
开源权重 Apache 2.0 · 官方 $0.834 输入 / $2.501 输出 每百万 token
已完成适配 newcclm-coding
GLM-5.3-FLASH

日常主力

320B 总参数 / 18B 激活 · 文本 / 图像 / 视频输入
官方 $0.15 输入 / $0.50 输出 每百万 token
已完成适配 newcclm · newcclm-fast
QWEN 3.8 FLASH

编码与长上下文

125B 主模型 + 51B N-gram 嵌入 / 6B 激活 · 开源权重
原生 26 万上下文，可扩到 100 万
已完成适配 newcclm-coding
LAGUNA S 2.1（免费）

零成本档

118B 总参数 / 8B 激活 · 26 万上下文 · 原生推理
免费模型 · 适合试用和批量任务
已完成适配 newcclm-free
DEEPSEEK V4 FLASH

备用线路

100 万 token 上下文 · 代码与逻辑推演
主线路故障时自动接管，无需配置
已完成适配 自动切换
About the rules of use
Each of these models has its own terms of service. When you use our service, you also need to follow each model's own usage policy — including not using it to generate unlawful content, and not reselling the model's capabilities.
If you're unsure whether a particular use is allowed, ask us first rather than trying it.
Will it get swapped without telling me? No. Choose "smart" and you get Qwen; choose "fast" and you get GLM. DeepSeek only takes over if the primary route fails — that's failover, not a downgrade.

How to start

Three steps, about two minutes. No technical configuration at all.

1
Open a terminal

macOS: search for "Terminal" in Spotlight.
Windows: press Win, type "Terminal" or "PowerShell", hit enter.
It's a black window. It looks unfamiliar, but all you do is paste two lines.

2
Paste this line and press enter

npx newcclm

3
Create an account, then just start chatting

Pick your own username and password. It's normal that the password doesn't show as you type. Every time after this, it's the same single command.

Common questions

The things people actually ask before signing up.

Do I need to understand any of this?
No. You copy one line, pick a username and password, and start typing. You never see an endpoint, a key, or a model name.
What's the difference between the two modes?
Fast is for email, rewriting, quick questions — you want the answer now. Smart is for complex analysis, long documents and code — slower, but it thinks further. You can switch any time by typing /fast or /smart.
What happens when the included usage runs out?
You're not cut off. You can top up a little and keep using it, or wait until next month. We never claim "unlimited" — you know your allowance before you pay.
Is there auto-renewal?
No. You buy 30 days; when it ends, it stops. We won't charge you again without you asking.
Which models do you use?
Three, all listed on this page: GLM-5.3-Flash for everyday work, Qwen3.8-Flash for coding, and DeepSeek V4-Flash as the backup route.
Is my data used for training?
No. We don't sell it or feed it into training. We only keep what billing needs — how much you used and when. If you need something stricter, talk to us.
What if it stops working?
Send us a screenshot. Most problems turn out to be one of a few common things and take a minute to fix. If it's on our side, we'll say so rather than blaming your setup.