Phi 4
by MicrosoftReleased Jan 10, 2025
microsoft/phi-4- Answers in a JSON schema
Microsoft Research Phi-4 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed. At 14 billion...
- Input / 1M tokens
- $0.084
- Output / 1M tokens
- $0.17
- Context
- 16K
- Longest answer
- 15K
What it costs
What each call is charged to the token's AI credit.
| Billed for | You pay |
|---|---|
| Input tokens | $0.084 / 1M |
| Output tokens | $0.17 / 1M |
A typical call
A short chat turn: 2,000 tokens in, 500 out.
- per call
- $0.0003
- calls per $20
- 79K
What it can do
From its listing. The router passes every field through.
- Reads imagesNo
- Reads files and PDFsNo
- Hears audioNo
- Calls toolsNo
- Answers in a JSON schemaYes
- Reasons before it answersNo
- Searches the webNo
- Caches long promptsNo
- Draws imagesNo
Call it
Any router key works, in the format your SDK already speaks.
import OpenAI from "openai";const ai = new OpenAI({ baseURL: "https://relayfor.si/api/v1", apiKey: process.env.RELAYFOR_ROUTER_KEY,});const reply = await ai.chat.completions.create({ model: "microsoft/phi-4", messages: [{ role: "user", content: "Welcome the 1,000th holder in one line." }],});Specifications
What its maker and providers list for it.
- Reads
- Text
- Writes
- Text
- Context
- 16,384 tokens
- Longest answer
- 14,745 tokens
- Intelligence index
- 5.9 (#208 of 220 rated)
- Knowledge cutoff
- Jun 30, 2024
- Released
- Jan 10, 2025
- Moderation
- None
Request fields
The fields its providers accept, passed through as you send them.
- frequency_penalty
- logit_bias
- max_tokens
- min_p
- presence_penalty
- repetition_penalty
- response_format
- seed
- stop
- structured_outputs
- temperature
- top_k
- top_p
Questions
How much does Phi 4 cost?
$0.084 per million input tokens and $0.17 per million output tokens, paid from the token's AI credit. A call of 2,000 tokens in and 500 out costs about $0.0003.
How do I call Phi 4?
Make a router key for one of your tokens on the dashboard, point any OpenAI or Anthropic SDK at https://relayfor.si/api/v1 and send the model id microsoft/phi-4. The token's AI credit pays for each call.
How long a prompt can Phi 4 take?
Up to 16,384 tokens of context, and up to 14,745 tokens in one answer.
Are my prompts stored?
No. The RFS Router passes each call through and keeps only its token counts and cost, for the token's usage page.
More from Microsoft
Output price per 1M tokens
Similar in price
Output price per 1M tokens
