openai
OpenAI: gpt-oss-20b
openai/gpt-oss-20b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
openai/gpt-oss-20b is an open-weight 21B-parameter text model from OpenAI released under Apache 2.0, using a Mixture-of-Experts design with 3.6B active parameters per pass. Its model card positions it for lower-latency, local, or specialized use and says MXFP4 quantization lets it run within 16GB of memory. Reasoning is mandatory with adjustable low, medium, or high effort, and the card reports function calling, browsing, code execution, and structured output support. OpenRouter's catalog lists a 131K-token context and a June 2024 knowledge cutoff.[1][2]
Strengths
- According to OpenAI's model card, reasoning effort is configurable across three levels, with function calling, web browsing, Python code execution, and structured outputs described as native agentic capabilities; OpenRouter's catalog also lists tool and structured-output parameters.[1][2]
- The card states the Apache 2.0 weights are fine-tunable on consumer hardware and that MXFP4 quantization lets the model run within 16GB of memory.[2]
- OpenRouter's catalog lists a 131K-token context window plus logprobs and seed parameters; the card highlights full chain-of-thought access for debugging, noted as not intended for end users.[1][2]
Limitations
- The model card warns the model was trained on the harmony response format and will not work correctly otherwise, so deployments must apply that format.[2]
- OpenRouter's catalog marks reasoning as mandatory rather than optional, which may matter for latency-sensitive use given the card's own latency guidance.[1][2]
- Evidence is a vendor card plus a catalog listing: performance figures come only from OpenRouter's catalog benchmarks, the listed modality is text-only, and real-world results should be validated on your own workloads.[1][2]
Where to use it
- A candidate for local or specialized deployments needing an open model that, per the card, fits in 16GB and can be fine-tuned; confirm on your own hardware.[2]
- Worth evaluating for agentic prototypes using function calling, browsing, or code execution as described by OpenAI, not as a proven result.[2]
- Potentially suited to structured-output or tool-calling pipelines with adjustable reasoning depth, per the supported parameters in OpenRouter's catalog.[1]
Sources
- [1] OpenRouter catalog · OpenAI: gpt-oss-20bSources checked: Nov 13, 2026
- [2] Model documentation · OpenAI: gpt-oss-20bSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for...
Specifications
- Model ID
- openai/gpt-oss-20b
- 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 $0.030 / out $0.130
- 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 | No measurement | 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 | Coding index | 20.7 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Agentic index | 1.2 | CC — below the B band | 13.11.2026 |
| Design Arena | models · datavizwin rate 39.7% · rank 127 | 933Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · websitewin rate 27.9% · rank 147 | 845Elo | 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) |
|---|---|---|---|---|---|---|
| Darkbloomfp8 | 131,072 | $0.018 | $0.090 | — | — | 100% |
| AkashMLfp4 | 131,072 | $0.018 | $0.090 | — | — | 100% |
| PrimeIntellectfp4 | 131,072 | $0.020 | $0.030 | — | — | 100% |
| DekaLLMbf16 | 131,072 | $0.029 | $0.140 | — | — | 100% |
| CoreWeavefp4 | 131,072 | $0.030 | $0.130 | — | — | 100% |
| DeepInfrabf16 | 131,072 | $0.030 | $0.140 | — | — | 100% |
| Parasailfp4 | 131,072 | $0.030 | $0.150 | — | — | 100% |
| SiliconFlowfp8 | 131,072 | $0.040 | $0.180 | — | — | 100% |
| Amazon Bedrock | 131,072 | $0.070 | $0.150 | — | — | 100% |
| Amazon Bedrock | 131,072 | $0.070 | $0.150 | — | — | 100% |
| Groq | 131,072 | $0.075 | $0.300 | — | — | 100% |
Provider data retrieved: 13.11.2026