MPT-30B
30B open model with 8K context that surpassed GPT-3, trained partly on H100s.
Retired
MPT-30B appears to be discontinued or shut down — legacy model line; MosaicML acquired by Databricks 2023 This page is kept as an archive record and is excluded from search engines.
About
MPT-30B, released by MosaicML on June 22, 2023, extended the MPT family with 30-billion-parameter Base, Instruct, and Chat models. Trained on 1 trillion tokens with an 8,000-token context window, it was designed to surpass the original GPT-3 on academic benchmarks at roughly one-sixth the parameter count, and was among the first publicly known LLMs trained in part on NVIDIA H100 GPUs.
The base model ships under Apache 2.0 and is commercially usable, with ALiBi for context extrapolation and FlashAttention for efficient inference. License caveat: the Instruct and Chat fine-tunes are under CC-BY-NC-SA, so commercial deployments should use the base weights or re-finetune. MosaicML was acquired by Databricks shortly after, and the MPT line fed into Databricks' open-model strategy.
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 81; Trending score 76/100 from researcher assessment.