Loading page
LightweightOpen SourceOpenAI
GPT-OSS-120B
GPT-OSS-120B is OpenAI's open-source lightweight model with 120 billion parameters, offering fast inference and a 128K token context window for developers.
Context 128K
Tier Lightweight
Knowledge Jan 2025
License Open Source
Input from
$0.037 / 1M tokens
across 12 providers
API Pricing
Cheapest on OpenRouter — 73% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.037 | $0.170 | - | |
| $0.050 | $0.250 | - | |
| $0.060 | $0.390 | - | |
| $0.070 | $0.300 | - | |
| $0.100 | $0.400 | - | |
| $0.150 | $0.600 | - | |
| $0.150 | $0.600 | - | |
| $0.150 | $0.600 | - | |
| $0.150 | $0.600 | - | |
| $0.150 | $0.600 | - | |
| $0.150 | $0.600 | - | |
| $0.174 | $0.697 | - | |
| $0.188 | $0.700 | $0.094 | |
| $0.350 | $0.750 | - |
Prices updated daily. Last check: Sep 6, 2026
GPT-OSS-120B pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
15.6 / 100
Coding
30.4 / 100
Math
93.4 / 100
Output Speed
181 t/s
Latency (TTFT)
463ms
Reasoning & Knowledge
- MMLU-Pro80.8%
- GPQA Diamond78.2%
- Humanity's Last Exam19.6%
Coding
- LiveCodeBench87.8%
- SciCode34.0%
Math
- AIME 202593.4%
Agentic & Tool Use
- Terminal-Bench Hard23.5%
- Terminal-Bench v2.126.2%
- τ²-bench65.8%
- τ-bench Banking12.8%
Instruction & Long Context
- IFBench69.0%
- Long-Context Reasoning52.0%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- OpenAI
- Family
- GPT-OSS
- Tier
- Lightweight
- Context Window
- 128K
- Knowledge Cutoff
- Jan 2025
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- Yes
- Subtypes
- Chat Completion
- Aliases
- GPT OSS Safeguard 120B
Strengths & Limitations
Strengths
- Open-source model weights available for local deployment and customization
- Fast inference speed at 207.02 output tokens per second
- 128K token context window supports long document processing
- Knowledge cutoff of January 1, 2025 provides recent training data
- Lightweight 120B parameter count balances performance with efficiency
- No API dependency required for inference
- Can be fine-tuned for domain-specific applications
Limitations
- No tool calling or function execution capabilities
- Text-only modality with no image or multimodal support
- Smaller parameter count than frontier models limits complex reasoning
- Requires local infrastructure and technical expertise to deploy
- Time to first token of 502ms slower than some specialized inference models
Key Features
•128K token context window
•Chat completion interface
•Open-source model weights
•Text-only input and output
•Streaming response capability
•Local deployment support
•Fine-tuning compatibility
•January 2025 knowledge cutoff
About GPT-OSS-120B
GPT-OSS-120B is OpenAI's open-source language model in the GPT-OSS family, positioned as a lightweight alternative to the company's flagship GPT models. With 120 billion parameters, this model represents OpenAI's entry into the open-source model space, making their technology accessible to developers who want to run models locally or modify them for specific use cases.
The model features a 128K token context window and focuses on text-only chat completion tasks. It delivers strong inference performance with 207.02 output tokens per second and a time to first token of 502 milliseconds. The model has a knowledge cutoff of January 1, 2025, providing relatively current training data. Unlike OpenAI's proprietary models, GPT-OSS-120B does not include tool calling capabilities, reflecting its streamlined design for core language tasks.
GPT-OSS-120B serves applications requiring fast, cost-effective text generation where the full capabilities of frontier models are unnecessary. Its open-source nature allows for local deployment, fine-tuning, and integration into custom applications without API dependencies, making it suitable for organizations with data privacy requirements or those seeking to reduce operational costs while maintaining quality text generation capabilities.
Common Use Cases
GPT-OSS-120B is well-suited for applications requiring fast, reliable text generation without the complexity of tool use or multimodal capabilities. Its open-source nature makes it ideal for organizations needing local deployment for data privacy, custom fine-tuning for domain-specific tasks, or integration into products without API dependencies. Common use cases include content generation, document summarization, customer service chatbots, code commenting, and batch text processing workflows where the 128K context window enables handling of long documents. The model's lightweight design and fast inference make it particularly valuable for high-throughput applications or resource-constrained environments where deploying larger frontier models would be impractical.
Frequently Asked Questions
How much does GPT-OSS-120B cost per million tokens?
GPT-OSS-120B pricing varies by provider and deployment method. Since it's open-source, you can also run it locally without per-token costs. Check the pricing table above for current rates across API providers.
What is GPT-OSS-120B best used for?
GPT-OSS-120B excels at text generation tasks requiring fast inference and long context handling, such as content creation, document summarization, and chatbot applications. Its open-source nature makes it ideal for organizations needing local deployment, custom fine-tuning, or applications with data privacy requirements.
How does GPT-OSS-120B compare to OpenAI's proprietary GPT models?
GPT-OSS-120B trades some advanced capabilities for accessibility and control. Unlike proprietary GPT models, it lacks tool calling and multimodal features but offers open-source weights for local deployment and customization. It's designed for use cases where fast, reliable text generation is needed without the full feature set of frontier models.