Expensive but Good Models
| AI Model | Company | Memory Size | Reading Cost* | Writing Cost* | Best For |
|---|---|---|---|---|---|
| Claude Sonnet 4.5 | Anthropic | 1M tokens | $3-6 | $15-22.50 | Complex projects, reliable results |
| GPT-5 | OpenAI | 400K tokens | $1.25 | $10 | Latest tech, three modes |
| Gemini 2.5 Pro | 1M+ tokens | TBD | TBD | Big projects, reading documents | |
| Qwen3 Coder | Multiple | 256K tokens | $0.20 | $0.80 | Writing code, open source |
Cheap Models
| AI Model | Company | Memory Size | Reading Cost* | Writing Cost* | Notes |
|---|---|---|---|---|---|
| DeepSeek V3 | DeepSeek | 128K tokens | $0.14 | $0.28 | Great value for everyday coding |
| DeepSeek R1 | DeepSeek | 128K tokens | $0.55 | $2.19 | Cheap but smart |
| Qwen3 32B | Multiple | 128K tokens | Varies | Varies | Open source, many providers |
| Z AI GLM 4.5 | Z AI | 128K tokens | TBD | TBD | Free license, smart reasoning |
Which One to Pick
| What Matters Most | Best AI | Why |
|---|---|---|
| Speed | Qwen3 Coder on Cerebras | Fastest |
| Quality | Claude Sonnet 4.5 | Most reliable for hard tasks |
| Value | DeepSeek V3 | Good quality, low cost |
How Reliable Are They?
Ranked from most to least reliable:- Claude Sonnet 4.5 - Almost always works correctly
- GPT-5 - Very good but sometimes has formatting issues
- Gemini 2.5 Pro - Good for normal tasks
- DeepSeek V3 - Reliable for simple tasks
- Qwen3 variants - Might need to try again for hard tasks
What Things Cost
| What You’re Doing | How Much Text | Claude Sonnet | DeepSeek V3 | Difference |
|---|---|---|---|---|
| Fix a small bug | 5K tokens | $0.05 | $0.001 | 50x cheaper |
| Add a feature | 50K tokens | $0.50 | $0.01 | 50x cheaper |
| Big code changes | 200K tokens | $2.00 | $0.04 | 50x cheaper |
Monthly Budget Examples
| Your Budget | Using Claude | Using DeepSeek | Smart Mix |
|---|---|---|---|
| $10/month | ~20 features | ~1000 features | Plan with DeepSeek, code with Claude |
| $50/month | ~100 features | ~5000 features | Important stuff: Claude, rest: DeepSeek |
| $100/month | ~200 features | ~10000 features | Hard tasks: Claude, easy tasks: DeepSeek |
Where to Get AI
| Provider | AI Models Available | How You Pay | Reliability | Help |
|---|---|---|---|---|
| CodinIT | Many | Buy credits | High | In the app |
| Anthropic | Claude only | Pay as you go | High | |
| OpenRouter | 100+ models | Pay as you go | High | Discord chat |
| OpenAI | GPT only | Pay as you go | High | Online forum |
| Local (Ollama) | Open source | Free | Depends on your computer | Community help |
Which Provider to Choose
Pick based on what you need:- Easy to use: CodinIT (no setup needed)
- Lots of choices: OpenRouter (access to all AI models)
- Direct: Anthropic or OpenAI (go straight to the source)
- Private: Ollama or LM Studio (runs on your computer)
What Others Are Using
What the CodinIT community uses:- See live stats at OpenRouter’s CodinIT page
- Most popular: Claude Sonnet 4.5 (40%)
- Growing fast: DeepSeek V3 (25%)
- Budget pick: Qwen3 variants (20%)