Model comparison

Phi 3 Mini 128k Instruct vs Phi 3 Mini 4k Instruct

Phi 3 Mini 128k Instruct is the stronger model overall, scoring 29.7 to 27.9 on the Noometry Index.

Last verified . 17 shared benchmarks.

Phi 3 Mini 128k Instruct Microsoft

29.7

Rank #305 Confirmed

Phi 3 Mini 4k Instruct Microsoft

27.9

Rank #328 Confirmed

Summary

  • They share 17 benchmarks with published results for both. Phi 3 Mini 128k Instruct scores higher in 4 categories and Phi 3 Mini 4k Instruct in 4 categories; 7 gaps are clear of the uncertainty.
  • The widest gap is in reasoning, where Phi 3 Mini 128k Instruct leads 19.6 to 14.1.

Side by side

Phi 3 Mini 128k Instruct and Phi 3 Mini 4k Instruct specifications
Phi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
ProviderMicrosoftMicrosoft
Noometry Index29.727.9
Released2024-04-232024-04-23
WeightsOpenOpen
Context window——
Max output——
Input $ / M tokens——
Output $ / M tokens——
Results tracked1935

Sponsored placements are available on pages like this one. Advertise on Noometry

Category by category

Coding Phi 3 Mini 128k Instruct leads

Phi 3 Mini 128k Instruct: 28.8 (#312), Phi 3 Mini 4k Instruct: 26.6 (#323)

Coding benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Coding10391093
BigCodeBench Instruct29.6%—
LiveBench Coding—15.5%
BigCodeBench Complete40.6%—
HumanEval+—59.1%
MBPP+—54.2%

Reasoning Phi 3 Mini 128k Instruct leads

Phi 3 Mini 128k Instruct: 19.6 (#256), Phi 3 Mini 4k Instruct: 14.1 (#328)

Reasoning benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Hard Prompts10281072
Chess Puzzles—0%
LiveBench Reasoning—26.8%
LiveBench Data Analysis—34.7%
Adversarial NLI—52.8%
BIG-Bench Hard—71.7%
HellaSwag—76.7%
LiveBench—22.4%
WinoGrande—70.8%

Math Phi 3 Mini 128k Instruct leads

Phi 3 Mini 128k Instruct: 31.6 (#222), Phi 3 Mini 4k Instruct: 26.6 (#257)

Math benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Math10891111
LiveBench Math—15.7%

Knowledge Phi 3 Mini 4k Instruct leads

Phi 3 Mini 128k Instruct: 26.8 (#254), Phi 3 Mini 4k Instruct: 28.5 (#246)

Knowledge benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Expert9841045
ARC (AI2) Challenge—84.9%
MMLU—68.8%
OpenBookQA—88%
TriviaQA—64%

Multilingual Phi 3 Mini 4k Instruct leads

Phi 3 Mini 128k Instruct: 25.2 (#285), Phi 3 Mini 4k Instruct: 26.3 (#280)

Multilingual benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Non-English10001021
LMArena Chinese10161021
LMArena French10391076
LMArena German10061044
LMArena Japanese899935
LMArena Korean856905
LMArena Russian10041022
LMArena Spanish10591085

Instruction Following Phi 3 Mini 128k Instruct leads

Phi 3 Mini 128k Instruct: 51.8 (#294), Phi 3 Mini 4k Instruct: 47.7 (#303)

Instruction Following benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Instruction Following10231053
LiveBench Instruction Following—39.1%

Long Context Phi 3 Mini 4k Instruct leads

Phi 3 Mini 128k Instruct: 30.4 (#289), Phi 3 Mini 4k Instruct: 31.7 (#276)

Long Context benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Longer Query9961044

Writing & Preference Too close to call

Phi 3 Mini 128k Instruct: 27.1 (#301), Phi 3 Mini 4k Instruct: 27.6 (#300)

Writing & Preference benchmarks
BenchmarkPhi 3 Mini 128k InstructPhi 3 Mini 4k Instruct
LMArena Text10501073
LMArena Creative Writing10241037
LMArena Multi-Turn9891018
LiveBench Language—9.2%

Frequently asked questions

Is Phi 3 Mini 128k Instruct better than Phi 3 Mini 4k Instruct?

Phi 3 Mini 128k Instruct is the stronger model overall, scoring 29.7 to 27.9 on the Noometry Index.

Is Phi 3 Mini 128k Instruct or Phi 3 Mini 4k Instruct better for coding?

Phi 3 Mini 128k Instruct scores higher on coding benchmarks: 28.8 versus 26.6 in the Noometry coding category.

How many benchmarks do Phi 3 Mini 128k Instruct and Phi 3 Mini 4k Instruct share?

17 benchmarks have published results for both models. Phi 3 Mini 128k Instruct has 19 scored results on Noometry and Phi 3 Mini 4k Instruct has 35.

Related comparisons

Go deeper