Back to Models
OpenAIOpen Source

GPT Opensources 120B

GPT Opensources 120B is a reasoning model developed by OpenAI. It has 117.0B parameters and a 128K context window, provided under an open license.

Parameters

117.0B

Context Window

128K

License

Apache 2.0

Release Date

2025-08-06

API Pricing

API pricing for this model is not yet available

Strengths

  • Equipped with advanced reasoning ability
  • 128K wide context understanding
  • Permissive Apache 2.0 license

Weaknesses

  • Gigantic file size of 240GB
  • High hardware requirements
  • Increased operational costs

Use Cases

  • Complex logical reasoning tasks
  • Large-scale document analysis
  • AI research in open source

Deep Analysis

Parameters

117B total / 5.1B active

MoE, 128 experts

AIME 2025

97.9%

With tools

GPQA Diamond

80.9%

With tools

SWE-Bench Verified

62.4%

With terminal

License

Apache 2.0

Release Date

August 5, 2025

Strengths

  • Fully open-weight Apache 2.0
  • 97.9% AIME 2025 with tools
  • Runs on single 80GB GPU (MXFP4)
  • Native agentic capabilities

Weaknesses

  • Text-only
  • Higher hallucination than o4-mini
  • Knowledge cutoff June 2024
  • Requires harmony format

Competitor Comparison

ModelSWEGPQAPrice
DeepSeek R1~55%~71%Free
Qwen 3.562.5%82.1%$0.50/$2.00
OpenAI o4-mini~63%$1.10/$4.40

GPT OSS 120B is OpenAI's first open-weight model (Apache 2.0). 117B MoE params, runs on single GPU. 97.9% AIME 2025 with tools.

Sources

Analysis generated: 2026-05-24