Introducing Caveman Compression & 4 Free Models
Today we're shipping two features that make AI Nexus even more cost-effective:
- Caveman Compression — automatically removes filler text from your API requests, saving 30-57% on tokens. Zero config.
- 4 Free Models — GLM-4.7-Flash, GLM-4V-Flash, CogView-3-Flash, and CogVideoX-Flash are now completely free.
🗜️ What Is Caveman Compression?
Caveman is a lightweight, rule-based compression engine that runs at the API gateway layer. Before your request reaches the upstream model, Caveman strips out text that adds tokens without adding meaning:
- Filler words: "thank you", "please", "sure", "basically" → removed
- Hedging: "I think", "it seems like", "probably" → removed
- Verbose phrasing: "due to the fact that" → "because"
- Redundant instructions: "make sure to", "it is important to" → removed
- Greetings & sign-offs: "Hi there", "hello", "thanks" → removed
Compression Ratios
| Language | Mode | Avg Savings | Example |
|---|---|---|---|
| 🇨🇳 Chinese | Lite | 57% | "你好，请问你能帮我写首诗吗？谢谢！" → "你能帮我写首诗吗？" |
| Standard (default) | 40% | Removes hedging + verbose phrasing | |
| Aggressive | 55% | Also removes filler adverbs and modal verbs | |
| 🇺🇸 English | Lite | 21% | "Thank you for your help! I think this is great." → "for your help! this is great." |
| Standard | 35% | Also compresses verbose phrases + redundant instructions |
How It Works
Caveman runs as a middleware proxy between your application and the AI model. It intercepts the messages[] array, applies rule-based compression per language, then forwards the optimized request:
Your App → [Caveman Proxy] → AI Model
↓
✂ Remove filler words
✂ Compress verbose phrases
✂ Strip greetings/sign-offs
✂ Auto-detect language (zh/en)
The engine is language-aware — it applies Chinese rules when it detects Chinese text, English rules for English, and intelligently skips short messages (<10 characters) where compression wouldn't help.
Is It Enabled?
Yes, it's on by default. Every request through https://www.tokencnn.com/v1 already benefits from compression. If you want to verify, check the response headers for X-Caveman-Compression: enabled.
You don't need to change a single line of code. Caveman works with any OpenAI SDK, any framework — it compresses at the proxy level, transparently.
🆓 4 Completely Free Models
We've partnered with Zhipu AI to offer the following models at $0/0 — no tokens consumed, no charges applied:
| Model | Type | Use Case |
|---|---|---|
| 🧠 GLM-4.7-Flash | Text | Chat, classification, content generation, RAG |
| 👁️ GLM-4V-Flash | Vision | Image understanding, OCR, visual QA |
| 🎨 CogView-3-Flash | Image Generation | Text-to-image, creative assets |
| 🎬 CogVideoX-Flash | Video Generation | Text-to-video, short clips |
📊 Real-World Savings Example
Say you're running a customer support chatbot that processes 50,000 conversations per month, averaging 500 tokens per conversation:
| Before Caveman | After Caveman | With Free Models | |
|---|---|---|---|
| Model | GPT-4o | DeepSeek V4 Flash | GLM-4.7-Flash |
| Tokens/month | 25M | ~15M (-40%) | $0 |
| Cost/month | $480 | $9 | $0 |
| Savings | — | 98% | 100% |
🔌 Get Started in 30 Seconds
- Sign up at tokencnn.com/register — no phone number needed, get $1 free credit
- Get your API key from the console
- Change one line in your code:
# Before client = OpenAI(base_url="https://api.openai.com/v1") # After client = OpenAI(base_url="https://www.tokencnn.com/v1")
That's it. Caveman compression is automatically active. Free models are available at model="glm-4.7-flash" — try them with zero spend.
🚀 Start Saving Today
No credit card required. Get $1 free credit + 4 free models on signup.
Get API Key →📚 What's Next
- More languages: Japanese and Korean compression rules coming next week
- Configurable levels: Choose between Lite / Standard / Aggressive per request
- Compression dashboard: See exactly how many tokens you saved in real-time