Basalt Labs Launches Monolith-1.0, a 1.57T-Parameter Open-Weight Reasoning Model
Basalt Labs on July 17 released Monolith-1.0, an open-source reasoning model with 1.57T parameters and 49.5B active parameters per token. The firm trained the architecture on 60T tokens of data across 12,288 Ascend 910C NPUs, setting the parameter limits to allow a 1M-token context window.
Weights, the tokenizer and evaluation tools carry an MIT license, allowing immediate public access. Basalt Labs reported benchmark scores of 95.9% on GPQA Diamond and 96.2% on MMLU-Pro, positioning the release as a top-tier open-weights model for frontier reasoning tasks.
From the sources (2 posts)
@teortaxestexshitposting as a post-training paradigm
@basaltlabsIntroducing Monolith-1.0: Frontier Reasoning, Open Weights - 1.57T-param Mixture-of-Experts, 49.5B active per token - Native 1M-token context (2²⁰), extended via a two-stage YaRN curriculum - Trained on 60T tokens across 12,288 Ascend 910C