openai
OpenAI: gpt-oss-120b
openai/gpt-oss-120b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
openai/gpt-oss-120b is an open-weight OpenAI text model described by its documentation as a 117B-parameter mixture-of-experts design with 5.1B parameters active per pass. The same documentation reports Apache 2.0 licensing, MXFP4 weights sized for a single 80GB GPU, configurable low/medium/high reasoning effort, full chain-of-thought access, fine-tuning, and agentic functions. OpenRouter lists text-only input and output, with reasoning mandatory and medium effort by default. These are source-reported capabilities; independent workload validation was not included.[1][2]
Strengths
- The model card reports Apache 2.0 open weights, commercial deployment permission, and full parameter fine-tuning, which may reduce adoption friction for custom products.[2]
- Its MoE design activates 5.1B of 117B parameters per pass, and the documentation says MXFP4 quantization allows operation on one 80GB GPU.[2]
- Configurable reasoning effort, inspectable chain-of-thought for debugging, function calling, browsing, Python execution, and structured outputs support agent-oriented workflows.[1][2]
Limitations
- OpenRouter lists only text inputs and outputs; the supplied evidence does not establish image, audio, or other modality support.[1]
- OpenAI's documentation warns that the model should only be used with the harmony response format and will not work correctly otherwise.[2]
- The catalog marks reasoning as mandatory, and the pack contains only vendor documentation plus catalog data; validate latency, cost, quality, and safety in the intended workload.[1][2]
Where to use it
- Consider it for customized commercial or self-hosted deployments when Apache 2.0 licensing and fine-tuning are important.[2]
- Evaluate it for agentic text applications that need function calling, web browsing, Python execution, or structured outputs.[1][2]
- It may fit production assistants where teams can exploit selectable reasoning effort and have suitable accelerator capacity.[2]
Sources
- [1] OpenRouter catalog · OpenAI: gpt-oss-120bSources checked: Nov 13, 2026
- [2] Model documentation · OpenAI: gpt-oss-120bSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized...
Specifications
- Model ID
- openai/gpt-oss-120b
- Creator
- openai
- Added to OpenRouter
- 05.08.2025
- Context length
- 131,072
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.150 / out $0.600
- Knowledge cutoff
- 30.06.2024
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | CC — below the B band | reasoning: mandatory · default medium · efforts: high, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Coding | CC — below the B band | — |
| Agents | CC — below the B band | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · default medium · efforts: high, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Intelligence index | 3.8 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Coding index | 30.4 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Agentic index | 3.7 | CC — below the B band | 13.11.2026 |
| Design Arena | models · 3dwin rate 29.4% · rank 129 | 884Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 33.4% · rank 138 | 953Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · datavizwin rate 43.6% · rank 122 | 985Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · gamedevwin rate 40.5% · rank 127 | 982Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 35.7% · rank 132 | 919Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · websitewin rate 32.5% · rank 140 | 961Elo | CC — below the B band | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Io Netfp4 | 131,052 | $0.030 | $0.150 | — | — | 76.67% |
| Near AIfp4 | 131,072 | $0.030 | $0.150 | — | — | 100% |
| Veniceint4 | 128,000 | $0.030 | $0.150 | — | — | 100% |
| CoreWeavefp4 | 131,072 | $0.030 | $0.170 | — | — | 100% |
| DekaLLMbf16 | 131,072 | $0.030 | $0.180 | — | — | 100% |
| AkashMLfp4 | 131,072 | $0.030 | $0.187 | — | — | 100% |
| DeepInfrabf16 | 131,072 | $0.037 | $0.170 | — | — | 100% |
| Crusoebf16 | 131,072 | $0.050 | $0.250 | — | — | 100% |
| Mancer 2fp8 | 131,072 | $0.055 | $0.250 | — | — | 100% |
| Sail Researchfp4 | 131,072 | $0.060 | $0.400 | — | — | 100% |
| DigitalOcean | 128,000 | $0.060 | $0.420 | — | — | 100% |
| PrimeIntellectfp4 | 131,072 | $0.061 | $0.130 | — | — | 100% |
| 131,072 | $0.090 | $0.360 | — | — | 62.07% | |
| BaseTenfp4 | 128,072 | $0.100 | $0.500 | — | — | 100% |
| BaseTenfp4 | 128,072 | $0.100 | $0.500 | — | — | 92.86% |
| Parasailfp4 | 131,072 | $0.100 | $0.750 | — | — | 100% |
| SambaNova | 131,072 | $0.140 | $0.950 | — | — | 100% |
| Amazon Bedrock | 131,072 | $0.150 | $0.600 | — | — | 100% |
| Nebiusfp4 | 131,072 | $0.150 | $0.600 | — | — | 100% |
| Amazon Bedrock | 131,072 | $0.150 | $0.600 | — | — | 100% |
| SiliconFlowfp8 | 131,072 | $0.150 | $0.600 | — | — | 100% |
| Together | 131,072 | $0.150 | $0.600 | — | — | 100% |
| Groq | 131,072 | $0.150 | $0.600 | — | — | 100% |
| Parasailfp4 | 131,072 | $0.150 | $0.750 | — | — | 100% |
| DeepInfrafp8 | 131,072 | $0.200 | $0.950 | — | — | — |
| Cerebrasfp16 | 131,072 | $0.350 | $0.750 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
- Amazon makes SGLang-Neuron publicly available for serving models on AWS TrainiumSGLang-Neuron, released by RadixArk with Annapurna Labs, is now publicly available via an AWS Deep Learning Container, enabling model serving and RL rollouts with SGLang on AWS Trainium chips.
- Four Amazon Bedrock request settings that change inference costsAn AWS walkthrough shows how cache duration, service tier, reasoning effort, and maxTokens change token counts and rates on Amazon Bedrock, with tests on Claude and Nova models.