Stockmark-2-100B-Instruct
100B from-scratch Japanese LLM under the MIT license.
About
Stockmark-2-100B-Instruct is Stockmark Inc.'s 100 billion parameter Japanese-focused language model, built from scratch on about 2 trillion tokens of English, Japanese, and code. Post-trained with SFT and DPO on synthetic Japanese data, it adds strong instruction following and a 32K context window over its beta release.
Notably released under the permissive MIT license and supported by Japan's GENIAC program, it is one of the largest openly licensed Japanese models, with strong Japanese MT-bench scores and NVIDIA NIM container availability.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
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: Verified via vendor site/docs/GitHub and reputable coverage, wave 131; Trending score 82/100 from researcher assessment.