Meta

Llama 3.2 90B Instruct

Multimodal
Zero-eval
#1InfographicsQA
#3VQAv2

by Meta

+
+
+
+
About

Llama 3.2 90B was developed as a flagship-tier open-source model, designed to provide advanced capabilities with 90 billion parameters. Built to serve applications requiring high-quality reasoning and generation, it represents a powerful option within the Llama 3.2 series for demanding tasks.

+
+
+
+
Pricing Range
Input (per 1M)$0.35 -$2.00
Output (per 1M)$0.40 -$2.00
Providers5
+
+
+
+
Timeline
AnnouncedSep 25, 2024
ReleasedSep 25, 2024
+
+
+
+
Specifications
Capabilities
Multimodal
+
+
+
+
License & Family
License
Llama 3.2
Performance Overview
Performance metrics and category breakdown

Overall Performance

13 benchmarks
Average Score
71.3%
Best Score
92.3%
High Performers (80%+)
5

Performance Metrics

Max Context Window
256.0K
Avg Throughput
54.6 tok/s
Avg Latency
1ms
+
+
+
+
All Benchmark Results for Llama 3.2 90B Instruct
Complete list of benchmark scores with detailed information
AI2D
multimodal
0.92
92.3%
Self-reported
DocVQA
multimodal
0.90
90.1%
Self-reported
MGSM
text
0.87
86.9%
Self-reported
MMLU
text
0.86
86.0%
Self-reported
ChartQA
multimodal
0.85
85.5%
Self-reported
VQAv2
multimodal
0.78
78.1%
Self-reported
TextVQA
multimodal
0.73
73.5%
Self-reported
MATH
text
0.68
68.0%
Self-reported
MMMU
multimodal
0.60
60.3%
Self-reported
MathVista
multimodal
0.57
57.3%
Self-reported
Showing 1 to 10 of 13 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+