GPT-OSS
Benchmarks
TaskDatasetSOTA resultTrendResults
GPT-OSS 20B
63.2Coverage
5
GPT-OSS 20B held-out (test)
34.56Perplexity
5
openai/gpt-oss-20b Long prompt
7.72TTFT (s)
3
openai/gpt-oss-20b Medium prompt
2.45Time To First Byte (s)
3
openai/gpt-oss-20b Short prompt
1.39TTFT (s)
3
gpt-oss-20b
2,656.61Throughput (tok/s)
2
GPT-OSS-20B workload
140,900Throughput (tokens/s)
2
Mini-GPT-OSS (val)
2.94Validation Loss
2
GPT-OSS Silo-Bench
6Effect (pp)
1