# DeepSeek V4 Flash Cost Analysis (May 2026)

Source: https://api-docs.deepseek.com/quick_start/pricing

## Pricing

| Metric | Cost per 1M tokens |
|--------|-------------------|
| Cached input | $0.0028 |
| Uncached input | $0.14 |
| Output | $0.28 |

## Real-world session costs (Ray's usage)

A heavy session with 39 API calls and 80K+ context per call:
- **Total per session**: ~$0.03
- **Cache hit rate**: 94%+ of input (2.5M cached vs 153K uncached)
- **Daily equivalent** (if used all day): ~$0.30-0.50
- **100 days of heavy use**: ~$3.37

## Why it's so cheap

1. **Context caching** — DeepSeek caches repeated input (conversation history). Cache hits cost 2% of miss price.
2. **V4 Flash is optimized for efficiency** — it's the "flash" tier, cheaper than the Pro tier.
3. **API calls are token-efficient** — most of the cost is input (conversation history), which caches heavily.

## Comparison

| Service | ~Cost per heavy session | Notes |
|---------|------------------------|-------|
| DeepSeek V4 Flash | $0.03 | Current default |
| Local Qwen3.6-35B | $0.00 | Free but slower on CPU offload |
| GPT-4.5 / Claude Opus | $5-15+ | Not used, just for scale comparison |