GPT OSS 120B
Open GPT reasoning model for self-hosted agents and controllable deployments
- Strongest at
- Math Top 22% of the past year's models · one benchmark so far
- Price per 1M tokens
- $0.15 / $0.6 input / output · OpenRouter
- Context window
- 131.1K tokens in · 32.8K out
Capability profile
Scored against 91 models released in the past year. 50 is the average model; further out is better.
No results yet: Reading & documents, Visual reasoning
Where these scores come from
Each capability combines the published benchmarks listed here, rescaled so 50 is the reference average and 15 points is one standard deviation. A hollow dot rests on a single benchmark.
- Coding Aider polyglot
- Writing EQ-Bench Creative Writing v3, EQ-Bench Longform Writing, Lech Mazur Writing comparison
- Reasoning SimpleBench, Mystery Game Puzzles
- Math OTIS Mock AIME 2024-2025
- Science GPQA diamond, WeirdML v2
- Computer use APEX-Agents, AutomationBench 1.0.6
Price against capability
The shaded corner holds anything better and cheaper. Prices blend input and output 3 to 1.
1 model from the past year scores higher on Math for less: Claude Haiku 5.5.
4 models from the past year score higher on Coding for less: DeepSeek V4.1 Flash, Claude Haiku 5.5, GPT-6 Luna and 1 more.
6 models from the past year score higher on Writing for less: DeepSeek V4.1 Flash, Gemma 4 31B IT, DeepSeek V4 Flash and 3 more.
7 models from the past year score higher on Reasoning for less: DeepSeek V4.1 Flash, Claude Haiku 5.5, DeepSeek V4 Flash 0731 and 4 more.
9 models from the past year score higher on Science for less: GPT-6 Luna, GLM-5.3-Flash, Claude Haiku 5.5 and 6 more.
5 models from the past year score higher on Computer use for less: Claude Haiku 5.5, GLM-5.3-Flash, DeepSeek V4.1 Flash and 2 more.
OpenRouter $0.15 / $0.6 · Cached input $0.075
The GPT OSS line
2 releases in the GPT OSS line.
Epoch benchmark boards
Best result on each board it has run. The tick marks the leader.
2 more boards
OpenAI's past year
24 release days in the past year, about one every 15 days. The next is expected around Nov 8, 2026. Forecasts ↗
Specifications
Specifications and prices from Models.dev; availability and limits may differ by provider. Benchmarks from Epoch AI (CC BY 4.0), LiveBench and the other sources named above.