GPT OSS 120B

An open-weight, 117B-parameter Mixture-of-Experts (MoE) language model designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized to run on a single H100 GPU with native MXFP4 quantization. The model supports configurable reasoning depth, full chain-of-thought access, and native tool use, including function calling, browsing, and structured output generation.

Reasoning
Tool Calling
Structured Output

Added Feb 3, 2026

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.030
Output: $0.17
Cache read: $0.030

Compare provider prices

Specifications

Context window: 128K
Max output: 16.4K
Parameters: 120B / 5.1B; Total / active
Avg output (7d): 294 tokens; Longer than 25% of models

Benchmarks

Artificial Analysis

LMArena

Vectara

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

11.6

Better than 46% of models compared

Coding Index

30.4

Better than 38% of models compared

Agentic Index

3.7

Better than 35% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

0.2%

Better than 3% of models compared

Harvey LAB-AA

Legal agentic work criterion pass rate

13.9%

Better than 0% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

0 Elo

Better than 3% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

612 Elo

Better than 30% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

4.0%

Better than 22% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

52.0%

Better than 45% of models compared

MLCR-AA

Medical long-context reasoning

1.1%

Better than 7% of models compared

Reasoning

HLE

Humanity's Last Exam

19.6%

Better than 67% of models compared

IFBench

Instruction-following benchmark

69.0%

Better than 83% of models compared

CritPt

Research-level physics reasoning

1.1%

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

34.0%

Better than 14% of models compared

LiveCodeBench

Contamination-free coding benchmark

87.8%

Better than 98% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

93.4%

Better than 94% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

80.8%

Better than 71% of models compared

AA-Omniscience Accuracy

Proportion of correctly answered questions

21.8%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

90.8%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

78.2%

Better than 64% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

23.5%

Better than 63% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

65.8%

Better than 60% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation

52.0%

Better than 45% of models compared

GDPval-AA (unversioned / legacy)

Economically valuable tasks

4.8%

Last updated Oct 3, 2026