Meta

Llama 3.3 70B Instruct

Zero-eval
#1BFCL v2
#2MBPP EvalPlus

by Meta

+
+
+
+
About

Llama 3.3 70B was introduced with refinements to the Llama 3 architecture, designed to incorporate improvements in instruction-following and task performance. Built to continue the evolution of Meta's 70B tier, it provides enhanced quality while maintaining the deployment characteristics valued by the open-source community.

+
+
+
+
Pricing Range
Input (per 1M)$0.20 -$0.89
Output (per 1M)$0.20 -$7.90
Providers9
+
+
+
+
Timeline
AnnouncedDec 6, 2024
ReleasedDec 6, 2024
+
+
+
+
Specifications
Training Tokens15.0T
+
+
+
+
License & Family
License
Llama 3.3 Community License Agreement
Performance Overview
Performance metrics and category breakdown

Overall Performance

9 benchmarks
Average Score
79.9%
Best Score
92.1%
High Performers (80%+)
5

Performance Metrics

Max Context Window
256.0K
Avg Throughput
451.9 tok/s
Avg Latency
1ms
+
+
+
+
All Benchmark Results for Llama 3.3 70B Instruct
Complete list of benchmark scores with detailed information
IFEval
text
0.92
92.1%
Self-reported
MGSM
text
0.91
91.1%
Self-reported
HumanEval
text
0.88
88.4%
Self-reported
MBPP EvalPlus
text
0.88
87.6%
Self-reported
MMLU
text
0.86
86.0%
Self-reported
BFCL v2
text
0.77
77.3%
Self-reported
MATH
text
0.77
77.0%
Self-reported
MMLU-Pro
text
0.69
68.9%
Self-reported
GPQA
text
0.51
50.5%
Self-reported
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+