Title: gpt-oss-20b vs Gpt Oss Safeguard 20B: Parametry, Licence %E2%80%94 LLM EU
Canonical: https://llmeu.com/cs/compare/gpt-oss-20b-vs-gpt-oss-safeguard-20b/
Markdown mirror: https://llmeu.com/cs/compare/gpt-oss-20b-vs-gpt-oss-safeguard-20b.md
Machine-readable: https://llmeu.com/llms.txt (index) - https://llmeu.com/llms-full.txt (full catalogue) - https://llmeu.com/llms-catalogue.json

# gpt-oss-20b vs Gpt Oss Safeguard 20B

Porovnejte dva a%C5%BE %C4%8Dty%C5%99i modely vedle sebe podle pol%C3%AD, kter%C3%A1 rozhoduj%C3%AD o nasazen%C3%AD: kontext, cena, region, licence, jazyky, hosting a co je ov%C4%9B%C5%99eno.

Tam, kde se dva modely li%C5%A1%C3%AD, tabulka uk%C3%A1%C5%BEe rozd%C3%ADl a skon%C4%8D%C3%AD. LLM EU nezve%C5%99ej%C5%88uje odznak v%C3%ADt%C4%9Bze bez dolo%C5%BEen%C3%A9ho sk%C3%B3re pro danou %C3%BAlohu a chyb%C4%9Bj%C3%ADc%C3%AD %C3%BAdaj z%C5%AFst%C3%A1v%C3%A1 chyb%C4%9Bj%C3%ADc%C3%AD, m%C3%ADsto aby se stal nulou.

## Kdy pou%C5%BE%C3%ADt

### [gpt-oss-20b](/cs/models/gpt-oss-20b/)

#### Siln%C3%A9 str%C3%A1nky

- runs in about 16GB of memory
- same 131,072-token window as the larger sibling
- Apache-2.0 licence
- reasoning-effort control

#### Omezen%C3%AD

- quality clearly below the 120B model
- harmony response format required
- no evaluation has been run by LLM EU

#### %C3%9Alohy

[Uva%C5%BEov%C3%A1n%C3%AD](/cs/tasks/reasoning/), [N%C3%ADzk%C3%A9 n%C3%A1klady](/cs/tasks/cheap/), [K%C3%B3dov%C3%A1n%C3%AD](/cs/tasks/coding/), [Vyhled%C3%A1v%C3%A1n%C3%AD (RAG)](/cs/tasks/rag/)

### [Gpt Oss Safeguard 20B](/cs/models/gpt-oss-safeguard-20b/)

#### Siln%C3%A9 str%C3%A1nky

- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- Apache-2.0 licence, so commercial self-hosting is permitted without a separate agreement
- instruction-tuned checkpoint with a chat template in the repository

#### Omezen%C3%AD

- catalogue only %E2%80%94 LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the published configuration records no context window, so the card carries none
- at 21.5B parameters, self-hosting needs a single accelerator with enough memory for the weights
- a safety classifier built on the gpt-oss architecture: it labels content against a policy supplied at inference time rather than answering questions

#### %C3%9Alohy

[Citliv%C3%A1 data](/cs/tasks/private/), [Z%C3%A1kaznick%C3%A1 podpora](/cs/tasks/customer-support/)

**%C3%9Alohy:** Neov%C4%9B%C5%99eno: poskytovatel toto tvrzen%C3%AD nedokumentoval.

## Porovnat modely

- [Command R (08-2024) vs gpt-oss-20b](/cs/compare/command-r-08-2024-vs-gpt-oss-20b/)
- [DeepSeek Coder V2 Lite Instruct vs gpt-oss-20b](/cs/compare/deepseek-coder-v2-lite-instruct-vs-gpt-oss-20b/)
- [DeepSeek R1 Distill Qwen 14B vs gpt-oss-20b](/cs/compare/deepseek-r1-distill-qwen-14b-vs-gpt-oss-20b/)
- [DeepSeek-R1-Distill-Qwen-32B vs gpt-oss-20b](/cs/compare/deepseek-r1-distill-qwen-32b-vs-gpt-oss-20b/)
- [DeepSeek V2 Lite vs gpt-oss-20b](/cs/compare/deepseek-v2-lite-vs-gpt-oss-20b/)
- [Devstral Small 2 (24B) vs gpt-oss-20b](/cs/compare/devstral-small-2-24b-vs-gpt-oss-20b/)