OpenModelsHQ

OpenAI

San Francisco, USA

OpenAI is based in San Francisco, USA. We track 15 open-weight models from it, and the newest is gpt-oss-safeguard-20b, released 18 Sep 2025.

Models tracked
15
LLM 3 · Speech 12
Newest release
gpt-oss-safeguard-20b
18 Sep 2025
Country
USA
Releases timeline
15 OpenAI releases from 26 Sep 2022 to 18 Sep 2025. The largest is gpt-oss-120b at 117B parameters. Dot area follows total parameters; a ring means the size is not published.
Releases timeline: data
ModelCategoryReleasedParameters
gpt-oss-safeguard-20bLLM18 Sep 202521.5B
gpt-oss-20bLLM4 Aug 202520.9B
gpt-oss-120bLLM4 Aug 2025117B
whisper-large-v3-turboSpeech1 Oct 2024809M
whisper-large-v3Speech7 Nov 20231.5B
whisper-large-v2Speech5 Dec 20221.5B
whisper-medium.enSpeech26 Sep 2022764M
whisper-small.enSpeech26 Sep 2022242M
whisper-base.enSpeech26 Sep 202273M
whisper-tiny.enSpeech26 Sep 202238M
whisper-largeSpeech26 Sep 20221.5B
whisper-mediumSpeech26 Sep 2022764M
whisper-smallSpeech26 Sep 2022242M
whisper-baseSpeech26 Sep 202273M
whisper-tinySpeech26 Sep 202238M

Origin

OpenAI was founded in December 2015 as a non-profit, with Elon Musk and Sam Altman as co-chairs, and is headquartered in San Francisco. Wikipedia says its founding charter stated an intention to collaborate openly, including by making some patents and research public, but that it later restricted access to its most capable models, citing competitive and safety concerns. A for-profit subsidiary was created in 2019. On 28 October 2025 OpenAI adopted a new structure: the for-profit business became OpenAI Group PBC, and the non-profit was renamed the OpenAI Foundation, which holds a 26% stake.

The flagship GPT models are proprietary. The open-weight releases covered here sit alongside them.

How it got here

OpenAI announced GPT-2 in February 2019 and at first released only limited versions, citing concerns about misuse. The full model came out in November 2019, according to Wikipedia.

Whisper, a speech recognition model, was released as open-source software in September 2022. Whisper Large V2 came on 8 December 2022 and Large V3 in November 2023. Large V3 Turbo, a faster version, appears on OpenAI’s Hugging Face page from October 2024.

The open-weight language models, gpt-oss, came on 5 August 2025. Wikipedia describes two reasoning models: gpt-oss-120b with about 117 billion parameters and gpt-oss-20b with about 21 billion. OpenAI says they are comparable to its o4-mini and o3-mini respectively. gpt-oss-20b is small enough to run on a device with more than 16 gigabytes of memory. In September 2025 OpenAI added gpt-oss-safeguard-20b, which its model card says interprets a written policy and gives reasoned decisions.

Licensing pattern

The gpt-oss models are Apache 2.0, according to both Wikipedia and the model card, which calls it a permissive licence with no copyleft restrictions or patent risk. The gpt-oss-safeguard card also lists Apache 2.0.

Whisper varies by version. The Large V3 card lists Apache 2.0, and the Large V3 Turbo card lists MIT.

These licences cover only the open releases named here. Wikipedia describes OpenAI’s flagship GPT models as proprietary.

Release lineage

Newest first
  1. gpt-oss-safeguard-20b21.5B
  2. gpt-oss-20b20.9B
  3. gpt-oss-120b117B
  4. whisper-large-v3-turbo809M
  5. whisper-large-v31.5B
  6. whisper-large-v21.5B
  7. whisper-medium.en764M
  8. whisper-small.en242M
  9. whisper-base.en73M
  10. whisper-tiny.en38M
  11. whisper-large1.5B
  12. whisper-medium764M

Largest first. The last line on each tile is the memory of the smallest device that fits the model at 4-bit.