Phi-4
Phi-4 is a model built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets.
text → text · made by Microsoft
Sold by 7 ways
| Seller | Lane | Rate |
|---|---|---|
| Microsoft Azure AI cloud | standard | $0.12per Mtok in$0.5per Mtok outmodels.dev · read 2026-09-16 |
| Requesty aggregatornot on the seller's list today | deepinfra flex | $0.05per Mtok in$0.1per Mtok outrouter.requesty.ai · read 2026-08-25 |
| Requesty aggregator | deepinfra | $0.07per Mtok in$0.14per Mtok out$0.07per Mtok cachedrouter.requesty.ai · read 2026-09-15 |
| DeepInfra aggregator | standard | $0.07per Mtok in$0.14per Mtok outapi.deepinfra.com · read 2026-08-25 |
| Hugging Face Inference Providers aggregator | deepinfra | $0.07per Mtok in$0.14per Mtok outrouter.huggingface.co · read 2026-08-25 |
| OpenRouter aggregator | standard | $0.07per Mtok in$0.14per Mtok outopenrouter.ai · read 2026-08-24 |
| ElectronHub aggregator | standard | $0.1per Mtok in$0.3per Mtok outapi.electronhub.ai · read 2026-09-16 |
Measured 12 standings
| Place | Board | Metric | Score |
|---|---|---|---|
| 4thof 105 | Vectara Hallucination Leaderboard | Hallucination Rate | 3.7 |
| 12thof 217 | Mmlu — Epoch AI | EM | 0.848 |
| 32ndof 36 | Balrog — Epoch AI | Average progress | 0.116 |
| 42ndof 54 | Live bench — Epoch AI | Global average | 41.61 |
| 45thof 49 | Lech mazur writing — Epoch AI | Mean score | 6.26 |
| 49thof 108 | Math level 5 — Epoch AI | mean_score | 64.936 |
| 190thof 222 | Chess puzzles — Epoch AI | mean_score | 1 |
| 214thof 313 | GPQA Diamond | mean_score | 56.061 |
| 236thof 290 | OTIS Mock AIME 2024-2025 | mean_score | 13.75 |
| 268thof 302 | Artificial Analysis · Intelligence Index | Artificial Analysis Intelligence Index | 6 |
| 306thof 402 | LMArena · Text | Score (Elo) | 1256 |
| 522ndof 656 | Epoch capabilities index — Epoch AI | ECI Score | 130.43 |
About
Phi-4 — a text model from Microsoft, sold by 6 companies from $0.07 in and $0.14 out per million tokens, placed 4th of 105 on Vectara Hallucination Leaderboard.
It takes text and returns text, with a context window of 16,384 tokens. It was published in December 2024, trained on material up to October 2023. The catalogue files it under reasoning. Six companies sell it. The cheapest is $0.07 in and $0.14 out per million tokens at DeepInfra. Beside the standard rate there is a separately routed lane. It has been measured on 12 boards, and stands best at 4th of 105 on Vectara Hallucination Leaderboard.
Every current figure
- Maker
- Microsoft
- Register
- model
- Takes
- text
- Returns
- text
- Context
- 16,384 tokens
- Longest answer
- 14,745 tokens
- Published
- December 2024
- Knowledge to
- October 2023
- Parameters
- 15 billion
- Licence
- not read
- Sellers
- 6
- Maker's own price
- not read
- Price
- $0.07 in and $0.14 out per million tokens — DeepInfra
- Boards
- 12
- Best place
- 4th of 105 — Vectara Hallucination Leaderboard
Known as 3 names
microsoft/phi-4deepinfra/microsoft/phi-4deepinfra/microsoft/phi-4:flex