Microsoft

Phi-3.5-mini-instruct

Zero-eval
#1Qasper
#1SQuALITY
#1QMSum
+11 more

by Microsoft

+
+
+
+
About

Phi-3.5 Mini was developed by Microsoft as a small language model designed to deliver impressive performance despite its compact size. Built with efficiency in mind, it demonstrates that capable language understanding and generation can be achieved with fewer parameters, making AI more accessible for edge and resource-constrained deployments.

+
+
+
+
Pricing Range
Input (per 1M)$0.10 -$0.10
Output (per 1M)$0.10 -$0.10
Providers1
+
+
+
+
Timeline
AnnouncedAug 23, 2024
ReleasedAug 23, 2024
+
+
+
+
Specifications
Training Tokens3.4T
+
+
+
+
License & Family
License
MIT
Performance Overview
Performance metrics and category breakdown

Overall Performance

31 benchmarks
Average Score
58.7%
Best Score
86.2%
High Performers (80%+)
4

Performance Metrics

Max Context Window
256.0K
Avg Throughput
23.0 tok/s
Avg Latency
1ms
+
+
+
+
All Benchmark Results for Phi-3.5-mini-instruct
Complete list of benchmark scores with detailed information
GSM8k
text
0.86
86.2%
Self-reported
ARC-C
text
0.85
84.6%
Self-reported
RULER
text
0.84
84.1%
Self-reported
PIQA
text
0.81
81.0%
Self-reported
OpenBookQA
text
0.79
79.2%
Self-reported
BoolQ
text
0.78
78.0%
Self-reported
RepoQA
text
0.77
77.0%
Self-reported
Social IQa
text
0.75
74.7%
Self-reported
MEGA XStoryCloze
text
0.73
73.5%
Self-reported
MBPP
text
0.70
69.6%
Self-reported
Showing 1 to 10 of 31 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+