LFM 3 1.7B

Overview

LFM 3 1.7B is a compact language model from Liquid AI, released on 2026-11-06 and made available through OpenRouter. Its headline feature is an unusually large 262,144-token context window for a model in the roughly 1.7-billion-parameter class. That combination makes it notable: it targets workloads where long-document comprehension, extended conversations, repository-scale text analysis, or multi-source synthesis matter, while retaining the latency and deployment advantages typically associated with smaller models.

Unlike many small models that are primarily optimized for short prompts or lightweight chat, LFM 3 1.7B is positioned for long-context text generation and reasoning. It is best understood as a small, efficient model designed to process far more input than most models of similar size, rather than as a direct replacement for frontier-scale systems.

Key capabilities and features

The model supports text-to-text generation, including summarization, drafting, extraction, classification, question answering, and reasoning over large bodies of text. The 262K-token context window enables use cases such as analyzing lengthy reports, comparing multiple policy documents, processing long meeting transcripts, or maintaining continuity across extended dialogue sessions.

Its reasoning capability is most valuable when paired with structured prompts and clearly scoped tasks. For example, it can identify contradictions across long documents, produce outlines from extensive source material, or answer questions that require referring to many sections of a large prompt. Because it is small, it may also be attractive for high-throughput applications where cost and responsiveness are important.

Technical specifications

• Provider: Liquid AI
• Model: LFM 3 1.7B
• Release date: 2026-11-06
• Parameter class: Approximately 1.7B, as indicated by the model name
• Context window: 262,144 tokens
• Modalities: Text input and text output
• Primary capabilities: Text generation, reasoning, long-context processing
• Availability: Listed on OpenRouter
• Max output tokens: Not specified in the provided model metadata; output limits may depend on OpenRouter configuration
• Pricing: Not specified here; users should consult the current OpenRouter model listing for live pricing
• Open weights/license: Not confirmed from the provided information; availability via OpenRouter does not necessarily imply open-weight release

Strengths and benefits

The main benefit of LFM 3 1.7B is its long-context efficiency. A 262K-token window gives developers and analysts the ability to include large inputs without relying as heavily on chunking, retrieval pipelines, or aggressive summarization. Its small size should make it more economical than larger long-context models for tasks that do not require top-tier reasoning depth.

It is also likely to be useful as a routing target in multi-model systems: use LFM 3 1.7B for broad context ingestion, extraction, or first-pass synthesis, then escalate difficult judgments to larger models when needed.

Limitations and caveats

The model’s small parameter count is also its main limitation. Compared with larger alternatives, it may be less reliable on complex multi-step reasoning, nuanced instruction following, advanced coding, mathematical problem solving, and tasks requiring deep world knowledge. Long context does not automatically guarantee perfect recall: models can still miss details, overemphasize recent information, or struggle with information buried in the middle of very large prompts.

Users should also verify practical constraints such as actual output-token caps, latency at maximum context, pricing, rate limits, and license terms before production use. For high-stakes domains, outputs should be validated with retrieval checks, citations, or human review.

Comparison

Relative to larger frontier models from providers such as Anthropic, Google, OpenAI, or larger Liquid AI models, LFM 3 1.7B should be viewed as a specialized efficient long-context model, not the most capable general-purpose reasoner. Compared with typical small LLMs, however, its 262K context window is a major differentiator and makes it unusually suitable for large-input workflows.