Compare · ModelsLive · 2 picked · head to head
phi-3-mini 3.8B vs phi-3-medium 14B
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
phi-3-medium 14B wins on 7/8 benchmarks
phi-3-medium 14B wins 7 of 8 shared benchmarks. Leads in knowledge · reasoning.
Category leads
knowledge·phi-3-medium 14Breasoning·phi-3-medium 14B
Hype vs Reality
Attention vs performance
phi-3-mini 3.8B
#36 by perf·no signal
phi-3-medium 14B
#48 by perf·no signal
Vendor risk
Who is behind the model
Microsoft
$3.00T·Big Tech
Microsoft
$3.00T·Big Tech
Head to head
8 benchmarks · 2 models
phi-3-mini 3.8Bphi-3-medium 14B
ANLI
phi-3-medium 14B leads by +4.5
ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations.
phi-3-mini 3.8B
29.2
phi-3-medium 14B
33.7
ARC AI2
phi-3-medium 14B leads by +8.9
AI2 Reasoning Challenge · tests grade-school level science knowledge with multiple-choice questions requiring reasoning beyond simple retrieval.
phi-3-mini 3.8B
79.9
phi-3-medium 14B
88.8
BBH
phi-3-medium 14B leads by +12.9
BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans.
phi-3-mini 3.8B
62.3
phi-3-medium 14B
75.2
HellaSwag
phi-3-medium 14B leads by +7.6
HellaSwag · tests commonsense reasoning by asking models to predict the most plausible continuation of everyday scenarios.
phi-3-mini 3.8B
68.9
phi-3-medium 14B
76.5
MMLU
phi-3-medium 14B leads by +12.3
Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge.
phi-3-mini 3.8B
58.4
phi-3-medium 14B
70.7
OpenBookQA
phi-3-mini 3.8B leads by +0.8
OpenBookQA · science questions that require combining a given core fact with broad common knowledge, mimicking an open-book exam setting.
phi-3-mini 3.8B
84.0
phi-3-medium 14B
83.2
TriviaQA
phi-3-medium 14B leads by +9.9
TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents.
phi-3-mini 3.8B
64.0
phi-3-medium 14B
73.9
Winogrande
phi-3-medium 14B leads by +21.4
WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs.
phi-3-mini 3.8B
41.6
phi-3-medium 14B
63.0
Full benchmark table
| Benchmark | phi-3-mini 3.8B | phi-3-medium 14B |
|---|---|---|
ANLI ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations. | 29.2 | 33.7 |
ARC AI2 AI2 Reasoning Challenge · tests grade-school level science knowledge with multiple-choice questions requiring reasoning beyond simple retrieval. | 79.9 | 88.8 |
BBH BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans. | 62.3 | 75.2 |
HellaSwag HellaSwag · tests commonsense reasoning by asking models to predict the most plausible continuation of everyday scenarios. | 68.9 | 76.5 |
MMLU Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge. | 58.4 | 70.7 |
OpenBookQA OpenBookQA · science questions that require combining a given core fact with broad common knowledge, mimicking an open-book exam setting. | 84.0 | 83.2 |
TriviaQA TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents. | 64.0 | 73.9 |
Winogrande WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs. | 41.6 | 63.0 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| — | — | — | — | |
| — | — | — | — |
People also compared