MosaicML
MosaicML built tools for training large language and diffusion models efficiently on commodity GPUs, releasing the MPT-7B and MPT-30B open-source models in 2023. Databricks acquired the company in July 2023 for roughly $1.3B, folding MosaicML's stack into the Databricks platform as the basis for what later became DBRX and the Mosaic AI training service. The standalone MosaicML brand was retired; the team and tech continue inside Databricks.
Acquisition Details
Databricks closed the $1.3B acquisition in July 2023 and folded MosaicML into its Mosaic AI brand. The original site redirects.
🩺 Health Signals
No checks have been run on this tool yet.
📅 Timeline
MosaicML launches with $37M Series A
https://techcrunch.com/2021/10/13/mosaicml-emerges-with-tools-to-make-training-machine-learning-models-faster/
Releases open-source MPT-7B model
https://www.databricks.com/blog/mpt-7b
Databricks acquires MosaicML for $1.3B
https://www.databricks.com/company/newsroom/press-releases/databricks-signs-definitive-agreement-acquire-mosaicml-leading-generative-ai
🔄 Alternatives to MosaicML
Hugging Face
🟢ActiveThe GitHub of machine learning — hosting models, datasets, and AI applications.
Cohere
🟢ActiveEnterprise AI platform providing NLP models for text generation, classification, and search.
Anthropic API
🟢ActiveAPI provider for Claude models, focused on AI safety research.
📰 Related Articles
Is MosaicML Still Active in 2026? (Status Explained)
MosaicML is no longer an independent entity, having been acquired by Databricks in July 2023. Its innovative AI development tools and team now operate under the Databricks Mosaic AI brand.
8 Best AI Training Platforms Like MosaicML in 2026
MosaicML — the LLM training-efficiency startup that built Composer, StreamingDataset, and the MPT foundation models — was acquired by Databricks for $1.3 billion in June 2023 and folded into Databricks Mosaic AI, leaving ML teams searching for a replacement or supplement in 2026. Databricks Mosaic AI is the direct heir with the same team and Apache-2.0 libraries, Together AI is best-in-class for serverless fine-tuning at $1.76/H100-hr, Anyscale wraps Ray for distributed multi-node training, Modal offers pythonic serverless GPUs for bursty workloads, Lambda Labs is the bare-metal H100 price anchor, Hugging Face TRL leads on RLHF and DPO, Axolotl delivers YAML-driven fine-tuning, and NVIDIA NeMo is the enterprise multi-node framework for owned GPU clusters. These eight tools like MosaicML, split cleanly between managed platforms and open-source frameworks and ranked by workload with a per-GPU-hour pricing chart, capability matrix, decision tree, and migration playbook, cover every reason a MosaicML user is searching for a new home in 2026.
🪦 Other dead data tools
❓ Frequently Asked Questions
Is MosaicML still a separate product?
No. Databricks completed the acquisition in 2023 and rebranded the offering as Mosaic AI inside the Databricks platform.
Are MPT models still maintained?
The MPT line was effectively replaced by Databricks' DBRX in 2024. MPT weights remain on Hugging Face but are not getting updates.
Can I still use the Composer library?
Yes. Composer and Streaming are open-source and actively maintained under the Mosaic GitHub org owned by Databricks.
Was MosaicML profitable when acquired?
Reported revenue at acquisition was modest (single-digit millions). Databricks bought the team and tech, not the revenue.