GPT OSS 120B High

GPT-OSS-120B High provides enhanced reasoning capabilities with high-effort thinking for complex problems. This variant offers deeper analysis and more thorough responses compared to the base model, making it ideal for challenging tasks that require extended reasoning. Activates 5.1B parameters per forward pass and is optimized to run on a single H100 GPU with native MXFP4 quantization.

Benchmark results

Benchmark Score Tags Source
AIME 2025 92.5% self-reported llm-stats link →
GPQA 80.9% self-reported llm-stats link →
IFBench 69.5% self-reported llm-stats link →
LiveCodeBench v6 81.9% self-reported llm-stats link →
MMLU-Pro 80.7% self-reported llm-stats link →
MMMLU 83.8% self-reported llm-stats link →
t2-bench 63.9% self-reported llm-stats link →