Microsoft

Phi 4

Type
language model
Context
128K tokens
Max output
4K
Released
11 December 2024
Knowledge cutoff
June 2024
API string
phi-4
Weights
open

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
Kilo Code $0.06 $0.14 31 Jul 2026
DeepInfra $0.07 $0.14 31 Jul 2026
OpenRouter $0.07 $0.14 31 Jul 2026
Azure Cognitive Services $0.125 $0.5 31 Jul 2026
Microsoft Foundry $0.125 $0.5 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Arena Score, mathematics 1,245.75 1,235–1,256
Arena Score, programming 1,231.82 1,222–1,242
Arena Score in Spanish 1,231.27 1,180–1,283
Arena Score in English 1,230.95 1,225–1,237
Arena Score in French 1,223.46 1,184–1,263
Arena Score, hard prompts 1,219.79 1,212–1,228
Arena Score, long queries 1,217.5 1,207–1,228
Arena Score, overall 1,216.84 1,212–1,221
Arena Score in Russian 1,207.8 1,197–1,219
Arena Score, multi-turn dialogue 1,205.9 1,196–1,216
Arena Score, expert questions 1,202.37 1,185–1,220
Arena Score, instruction following 1,201.73 1,195–1,208
Arena Score, creative writing 1,182.03 1,172–1,192
MATH, difficulty level five 64.94 % · with a tuned harness Epoch evaluations
Creative writing (Lech Mazur’s evaluation) 62.6 % lechmazur/writing Github repository
GPQA Diamond — graduate-level questions 41.41 % · with a tuned harness Epoch evaluations
Mock AIME 2024–2025 — olympiad problems 13.66 % · with a tuned harness Epoch evaluations
BALROG — game environments 11.6 % Balrog Leaderboard