Model Tracker
M2M-100
M2M-100 is an AI model made by Meta. Meta released it on 19 October 2020. It takes text and returns text. Its weights are open.
- Released
- 19 October 2020 — ai.meta.com, read 16 October 2026
- Status
- Not confirmed: The maker's pages found do not state that M2M-100 is preview, beta, experimental, active, stable, generally available, or retired. Its published checkpoints alone do not establish a status under the register's rules.
- Takes
- Text — ai.meta.com, read 16 October 2026
- Returns
- Text — ai.meta.com, read 16 October 2026
- Open weights
- Yes — ai.meta.com, read 16 October 2026
- Licence
- Not confirmed: The maker's M2M-100 announcement and model repository pages found do not name a licence for the model weights.
- Context window
- Not confirmed: The maker's pages found do not print a standard context-window limit for M2M-100.
- Longest answer
- Not confirmed: The maker's pages found do not print a standard maximum output-token limit for M2M-100.
- Knowledge cutoff
- Not confirmed: The maker's pages found do not state a knowledge-cutoff date for M2M-100.
- API price, input
- Not confirmed: No standard-tier API input price for M2M-100 is printed on the maker's pages found.
- API price, output
- Not confirmed: No standard-tier API output price for M2M-100 is printed on the maker's pages found.
- Where
- Downloadable weights — github.com, read 16 October 2026
- Parameters
- 15,000,000,000 — ai.meta.com, read 16 October 2026
- Model card
- Not confirmed: No maker-published M2M-100 model card or system card was found; the repository page is a usage README, not a card identified as such.
Questions
When was M2M-100 released?
Meta released M2M-100 on 19 October 2020.
Are the weights of M2M-100 open?
Yes: Meta publishes the weights of M2M-100 for download.
What does M2M-100 take and return?
M2M-100 takes text and returns text.
How many parameters does M2M-100 have?
M2M-100 has 15,000,000,000 parameters, as Meta publishes it.
More from Meta
- Audiobox
- AudioGen
- Chameleon
- CM3leon
- Code Llama
- DINOv2
- DINOv3
- Emu
- Emu Edit
- Emu Video
- EnCodec
- I-JEPA
- ImageBind
- LLaMA
- Llama 2
- Llama 3
- Llama 3.1
- Llama 3.2
- Llama 3.3
- Llama 4 Maverick
- Llama 4 Scout
- Make-A-Video
- Movie Gen
- Multi-Band Diffusion
Sources
- The first AI model that translates 100 languages without relying on English data (read 16 October 2026)
- fairseq/examples/m2m_100/README.md (read 16 October 2026)