Moore Threads open-sources MusaCoder code LLM, 9B/27B parameters trained end-to-end on domestic GPUs
Original title:摩尔线程开源 MusaCoder:首个基于国产全功能 GPU 全栈训练的代码大模型
AI summary
Moore Threads has released and open-sourced MusaCoder, a code large language model in 9B and 27B parameter sizes, the industry's first open-source model to complete full-chain training and validation on a domestic GPU computing base. The post-training pipeline was completed on a KUAE intelligent computing cluster based on MTT S5000, and supports automatic generation of high-performance native CUDA/MUSA kernel code from standard PyTorch operators. On the KernelBench benchmark, MusaCoder-27B-RL surpassed mainstream SOTA code models including Claude Opus, DeepSeek-V4 Pro, GLM-5.1 and Kimi K2.6 with an Overall Pass@8 of 93.2% and Avg.@8 of 88.60%.
Key Takeaways
- Moore Threads has released and open-sourced MusaCoder,a code large language model in 9B and 27B…
- The post-training pipeline was completed on a KUAE intelligent computing cluster based on MTT…
- On the KernelBench benchmark,MusaCoder-27B-RL surpassed mainstream SOTA code models including…
- 1 and Kimi K2.
- 6 with an Overall Pass@8 of 93.
Quick Facts
- Published
- Updated
- Source
- IT之家·人工智能
- Category
- Model Releases
- AI score
- 72
摩尔线程这个模型直接瞄准GPU底层算子,KernelBench分数压过了不少主流大模型,虽然场景窄,但在国产硬件自建软件生态的路上,迈出了挺关键一步。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Moore Threads open-sources MusaCoder code LLM, 9B/27B parameters trained end-to-end on domestic GPUs. June 10, 2026. https://new.nat6.net/en/news/cmq7t3hhr01f3slepkahaev5l/