LiquidAI: LFM2.5-2.6B (free)

LiquidAI: LFM2.5-2.6B (free) is a compact reasoning model from Liquid AI. It sits in the text category and is distributed on api14 under the model id liquid/lfm-2.5-2.6b:free. Liquid AI positions it as a small model built for reasoning-heavy tasks rather than raw size, and describes it as suited to agent workflows, data extraction, retrieval augmented generation, and long-context processing.

The model takes text in and returns text out, with a context window of 65,536 tokens and a maximum output of 8,192 tokens. Its supported parameters include temperature, top_p, top_k, min_p, frequency_penalty, presence_penalty, repetition_penalty, seed, stop, logprobs, top_logprobs, max_tokens, and max_completion_tokens. It also accepts reasoning and include_reasoning, plus response_format and structured_outputs for constrained generation, and tools and tool_choice for function calling. The combination of a 64K context window and an 8K output ceiling makes it practical for inputs that are long but where the answer itself stays bounded.

Those capabilities point to a fairly specific set of uses. Agent workflows are a natural fit because the model supports tool calling and structured outputs, so it can pick tools, return arguments in a schema, and iterate over steps. Data extraction works the same way: feed in a document or record set and request JSON or another structured shape via response_format or structured_outputs. For RAG, the long context lets you pack retrieved passages alongside the question without aggressive trimming. Long-context processing more generally covers summarisation, classification, and question answering over large inputs, while the reasoning parameter lets you surface intermediate reasoning when you want it, and the logprobs family supports scoring and confidence checks. Liquid AI advises against using it for agentic coding, so keep coding agents on a different model.

Because api14 exposes an OpenAI-compatible API, you call this model with the model id liquid/lfm-2.5-2.6b:free using the same request shape you already use for OpenAI models. Point your client at the api14 base URL, set the model field to liquid/lfm-2.5-2.6b:free, and send your messages. Sampling controls (temperature, top_p, top_k, min_p, the three penalties, seed), output limits (max_tokens, max_completion_tokens), stop sequences, structured output options (response_format, structured_outputs), tool calling (tools, tool_choice), reasoning controls (reasoning, include_reasoning), and log probability options (logprobs, top_logprobs) are all available as request parameters. The model is free on api14, so you can evaluate it without a paid commitment, though the free listing may come with its own availability or rate conditions.

Two limits are worth keeping in mind when you choose. The 8,192 token maximum output bounds how much the model can produce in one response, so work that needs very long single outputs, such as generating a whole large file, will need chunking or a model with a higher output ceiling. And the text-only input and output means there is no image or audio handling here; if your task involves other modalities, another model will fit better.