OpenAI

GPT OSS 20B

Zero-eval
#2CodeForces
#2HealthBench
#2HealthBench Hard

by OpenAI

+
+
+
+
About

GPT-OSS 20B was created as a smaller open-source variant, designed to balance capability with accessibility for researchers and developers. Built with 20 billion parameters, it provides a more approachable entry point for those seeking to work with open-source language models while benefiting from OpenAI's architectural insights.

+
+
+
+
Pricing Range
Input (per 1M)$0.05 -$0.10
Output (per 1M)$0.20 -$0.50
Providers4
+
+
+
+
Timeline
AnnouncedAug 5, 2025
ReleasedAug 5, 2025
+
+
+
+
License & Family
License
Apache 2.0
Performance Overview
Performance metrics and category breakdown

Overall Performance

7 benchmarks
Average Score
52.3%
Best Score
85.3%
High Performers (80%+)
2

Performance Metrics

Max Context Window
262.1K
Avg Throughput
705.0 tok/s
Avg Latency
2ms
+
+
+
+
All Benchmark Results for GPT OSS 20B
Complete list of benchmark scores with detailed information
MMLU
text
0.85
85.3%
Self-reported
CodeForces
text
0.84
83.9%
Self-reported
GPQA
text
0.71
71.5%
Self-reported
TAU-bench Retail
text
0.55
54.8%
Self-reported
HealthBench
text
0.42
42.5%
Self-reported
Humanity's Last Exam
multimodal
0.17
17.3%
Self-reported
HealthBench Hard
text
0.11
10.8%
Self-reported
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+