Basalt<br>Try Monolith
Monolith-1.0 · Released July 17, 2026<br>Monolith.<br>The frontier is open. Monolith-1.0 leads every major reasoning benchmark, and its weights are on Hugging Face.<br>Chat on basaltlabs.org
First place,<br>on everything that matters.<br>Reasoning, math, science, and long-context analysis. Evaluation protocols and reproduction details are published alongside every release.
Benchmark<br>Monolith-1<br>GPT-5.4<br>Claude Opus 4.6<br>Gemini 3.1 Pro<br>Kimi K2.6
Humanity's Last Exam<br>99.4%<br>40.0%<br>40.0%<br>44.4%<br>34.7%
AIME 2025<br>100%<br>99.2%<br>96.7%<br>98.3%<br>96.4%
GPQA Diamond<br>95.9%<br>92.8%<br>91.3%<br>94.3%<br>90.5%
MMLU-Pro<br>96.2%*<br>90.2%<br>89.7%<br>90.8%<br>88.9%
* Evaluated using Basalt's internal harness. Full methodology in the technical report.
Model card<br>Specifications.<br>Full model card, evaluations, and training details available in the paper.
Parameters<br>1.6T total / 49B active
Architecture<br>Mixture of Experts, 128 experts
Context window<br>1,048,576 tokens
Training compute<br>1.8 × 10²⁵ FLOPs
Training data cutoff<br>December 2025
License<br>MIT
Availability<br>Two ways to access Monolith.<br>basaltlabs.org<br>Free chat interface. Login with any email.
Weights<br>Full 1.6T weights on Hugging Face. No waitlist.