Phi 4 — Benchmarks

Benchmark scores for Phi 4 aggregated from public leaderboards (7 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

83.9

90% range 79.2–88.6 · from 4 boards

CodingNot enough data yet
AgentsNot enough data yet
MathNot enough data yet
ScienceNot enough data yet
ReasoningNot enough data yet

Phi 4 reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

GPQA Diamond
GPQA Diamond is a set of extremely hard, graduate-level science questions (physics, chemistry, biology) written by domain experts and filtered so that skilled non-experts with web access still fail. It measures genuine reasoning rather than memorization.
56.1%
56.1%
median 54.0%open leader 93.1%
Overall best: Gemini 4 Argon (high) 96.4%
53rd percentile · #46 of 98 open · #212 of 300 all
#46 of 98
#212 of 300
Independent
Epoch AI · Dec 2024
BALROG
BALROG (Benchmarking Agentic LLM and VLM Reasoning On Games) tests long-horizon agentic reasoning across six game environments — BabyAI, Crafter, TextWorld, Baba Is AI, MiniHack and NetHack — scoring the average progress a model makes toward completing each game. It was built by UCL's DARK lab and collaborators.
11.6%
11.6%
median 16.2%open leader 34.9%
Overall best: GPT 6 Astra Max 68.3%
29th percentile · #11 of 15 open · #36 of 40 all
#11 of 15
#36 of 40
Independent
Epoch AI · Dec 2024

Phi 4 math benchmarks

Symbolic and quantitative problem-solving, from competition math to multi-step calculation.

AIME 2024/2025
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
13.8%
13.8%
median 52.8%open leader 98.6%
Overall best: Claude Fable 5 (high) 100.0%
34th percentile · #57 of 86 open · #231 of 283 all
#57 of 86
#231 of 283
Independent
Epoch AI · Dec 2024
MATH Level 5
MATH Level 5 covers the hardest tier of competition-mathematics problems, testing multi-step symbolic and quantitative reasoning.
64.9%
64.9%
median 41.6%open leader 96.6%
Overall best: GPT 5 (Aug 07, 2025, high) 98.1%
79th percentile · #9 of 39 open · #49 of 107 all
#9 of 39
#49 of 107
Independent
Epoch AI · Dec 2024

Phi 4 knowledge benchmarks

Breadth and depth of factual knowledge across many subjects, paired with the reasoning to apply it.

MMLU-Pro
MMLU-Pro is a harder, cleaned-up successor to MMLU with ten answer choices and more reasoning-heavy questions across 14 subjects, measuring broad knowledge and reasoning together.
70.4%
70.4%
median 52.1%open leader 88.0%
Overall best: Gemini-3.1-Pro 91.2%
75th percentile · #37 of 145 open · #98 of 257 all
#37 of 145
#98 of 257
Independent
TIGER-Lab

Other results (1)· leaderboard-only, not part of the llmrun Score

MMLU
MMLU (Massive Multitask Language Understanding) spans 57 subjects from history to law to medicine as multiple-choice questions. It is the long-standing default for broad knowledge and remains the most widely-reported general benchmark.
84.8%
84.8%
median 62.5%open leader 87.2%
Overall best: GPT 4o (Nov 20, 2024) 88.1%
95th percentile · #5 of 85 open · #11 of 136 all
#5 of 85
#11 of 136
Lab-reported
Epoch AI · Dec 2024

Phi 4 instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1256.3
1256.3
median 1321.5open leader 1487.1
Overall best: Gemini 4 Argon (High) 1523.9
32nd percentile · #128 of 187 open · #310 of 405 all
#128 of 187
#310 of 405
Arena
Arena · Oct 2026

Score vs VRAM in the 24 GB tier

Phi 4
Other models
Best Score for its VRAM
x: VRAM (GB) · y: llmrun Score

#22 of 35 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Phi 4 vs its neighbours

Same boards for the nearest open models per llmrun Score and the best model that fits 24 GB. The highest score in each row is marked.

Board	Phi 4this model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	Mistral Small 3.2 24B Instruct 2506nearest by ScoreQ4_K_M · 18 GB	Magistral Small 2509nearest by ScoreQ4_K_M · 18 GB	Qwen2.5 Coder 32B Instructnearest by ScoreQ4_K_S · 23 GB
llmrun Score	83.9	122.1	85.7	85.6	81.9
GPQA Diamond	56.1%	—	49.0%	47.6%	—
AIME 2024/2025	13.8%	—	30.3%	28.1%	—
LMArena Text	1256.3	1437.8	1356.1	—	1270.6
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 5 boards, Arena: 1 board, TIGER-Lab: 1 board) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.