Liquid AI review

Free credits: no, and the models cost nothing to download, run and fine-tune, including commercially under a stated annual revenue cap. The catalogue runs LFM2.5 2.6B, LFM2.5 1.2B Instruct and LFM2.5 1.2B Thinking, so prototyping carries no token bill.

The largest context window across the library is 128,000. That sits well short of the 1,000,000 site median for the biggest window, so long-document work needs chunking or a different provider. The published model list shows no longer variant.

Liquid AI is Liquid AI, Inc., a US company founded 2023 and headquartered in the United States. The models are its own LFM line rather than resold third-party weights, so licensing questions go to the maker rather than to a gateway.

The library covers text, vision, audio, embedding, extraction and translation tasks: LFM2.5 2.6B, LFM2.5 1.2B Instruct and LFM2.5 1.2B Thinking. The set mixes small checkpoints with larger mixture models, and Japanese variants sit alongside English ones for transcription and PII work.

Support runs through email, help centre and community. That covers written answers and a help centre, with a community space for peer advice. Live chat and phone are not listed, so time-sensitive production issues need your own escalation path.

Yes within limits: the models are free to download, run and fine-tune, including commercially, though the free terms apply under a stated annual revenue cap. The counterparty on those terms is Liquid AI, Inc., and above the cap you need a separate arrangement.

Narrower. Azure AI Foundry lists gpt 6 astra, gpt 6 luna and gpt 6 sol behind a metered endpoint, chat, coding, embedding, image and video included, where Liquid AI instead hands you weights to run yourself.

When variety matters more than running weights yourself. Supascale AI fronts DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra from many vendors, while Liquid AI keeps one LFM line. Choose Liquid AI for control, the gateway for range.