AI Model Atlas / Microsoft
Phi 4
One million input tokens cost $0.070 and one million output tokens $0.14. Context window: 16k tokens, about 27 A4 pages.
What it costs in practice
Pricing per million tokens doesn't mean anything to anyone, so I convert it to three tasks that people actually give to models. The usage assumptions are listed for each row so you can verify them.
| Task | Input | Output | Cost |
|---|---|---|---|
| 1500-word articlebrief plus finished text in English | 1,500 tokens | 2,200 tokens | $0.0004 |
| 100 chat queries700 tokens per question and 400 per answer | 70,000 tokens | 40,000 tokens | $0.01 |
| Hour of work with an agentagent reads repository and writes fixes | 250,000 tokens | 30,000 tokens | $0.02 |
Technical details
| Provider | Microsoft |
| Modalities | text → text |
| Context window | 16k tokens (~27 A4 pages) |
| Maximum response | 15k tokens |
| Step-by-step reasoning | no |
| Tool calling | no |
| Forced JSON format | yes |
| Knowledge cutoff | 2024-06-30 |
| Cache read | no data |
| Content moderation | no extra filter |
How it compares to the rest
The database contains 469 models. In terms of input pricing Phi 4 is more expensive than 11 percent of paid models. A longer context window is available in 457 models.
Input price history
2026-09-08 · $0.0702026-11-11 · $0.070
The chart shows the price of a million input tokens on consecutive measurement days. History starts on the day I launched the atlas and grows with each day. I don't backfill data.
Alternatives at a similar price
Models from other providers whose cost for the same article comes out closest.
| Model | Input / M | Output / M | 1500-word article | Context |
|---|---|---|---|---|
| Ternary Bonsai 2 9BPrism ml | $0.080 | $0.13 | $0.0004 | 262k |
| Gemma 3 12BGoogle | $0.050 | $0.15 | $0.0004 | 131k |
| DeepSeek V4 Flash 0731 (batch)DeepSeek | $0.048 | $0.14 | $0.0004 | 1M |
| Command R7B (12-2024)Cohere | $0.038 | $0.15 | $0.0004 | 128k |
Compare Phi 4 with another model or go back to the full database 469 models.