M2M100
Many-to-many NMT translating directly across all 9,900 directions of 100 languages, trained without English-centric pivoting.
About
M2M100 is Meta's landmark many-to-many translation model from the 2020 paper on moving beyond English-centric multilingual MT. Instead of routing everything through English, it trains one encoder-decoder model on mined parallel data across 100 languages, so any pair translates directly. It ships in 418M, 1.2B, and 12B sizes, and the paper's headline claim is double-digit BLEU gains on low-resource pairs versus English-pivot baselines.
The model is MIT licensed and remains a workhorse baseline in MT research, with wide third-party support including GGUF conversions and FastAPI serving templates. It runs through the standard Hugging Face transformers pipeline with forced language-ID decoding. Newer models cover more languages, but few match its combination of openness, size options, and direct any-to-any translation.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (paper): up to +10 BLEU over English-centric baselines on low-resource pairs across the 100-language set
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Meta FAIR paper (2020) with mined-data training recipe confirmed via Hugging Face; 418M/1.2B/12B checkpoints MIT licensed with broad ecosystem support (GGUF, serving templates). Strengths: permissive MIT license, 9,900 direct translation directions, still a standard research baseline. Docked for: (1) quality lags NLLB-200 and modern LLM translators; (2) 100-language coverage is now mid-tier. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Meta FAIR 2020 paper; 418M/1.2B/12B checkpoints
- · Direct any-to-any translation across 100 languages (9,900 directions)
- · Wide ecosystem: GGUF conversions, serving templates, tutorials