LFM 3 1.7B
Neural Network
LFM 3 1.7B is a compact text model with a 256K token context, fine-tuned for agents and tool calling.
Specifications
- Input
- text, documents
- Output
- text
- Context size
- 262 144 tokens
- Max answer length
- 8 192 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works LFM 3 1.7B?
LFM 3 1.7B is a compact 1.7B parameter text model with a 256K token context, fine-tuned for agent scenarios, tool calling, and structured output. It performs reasoning before answering, making it reliable for tasks where step-by-step logic is more important than instant response. Its small size makes it ideal for streaming tasks with predictable costs: long documents fit entirely in the window, and responses can be received as JSON for downstream processing. Via BotHub, the model is accessible from Russia without a VPN or foreign card, with pay-as-you-go billing in rubles and no subscription—tokens do not expire. It is available alongside other models, making it easy to compare results with larger variants and switch without rewriting integrations. A unified OpenAI-compatible API helps integrate LFM 3 1.7B into multi-model systems. Companies have access to contracts, invoices, EDI, admin panels, and limits. Scenarios: agents and tool-calling chains, parsing long contracts, extracting fields into fixed schemas, code refactoring, and product assistants.Frequently asked questions about LFM 3 1.7B
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is a compact 1.7B parameter Liquid AI model: it answers questions, parses documents, reasons before answering, and calls functions. A context of up to 262,144 tokens allows working with long materials. Suitable for summarization, data extraction, simple assistants, and scenarios where speed and budget efficiency are important.
A single response is limited to 8,192 output tokens, while the input context reaches 262,144 tokens. How many pages this covers depends on the language and formatting. If longer material is needed, it is logical to split it into several requests and assemble the result in parts.
Yes, reasoning before answering is a supported capability: it can go through the steps of a task first and then provide the result. This is useful where the solution process is important—analyzing conditions, checking calculations, or logic puzzles. You define the format and detail level in your prompt.
Yes, both are supported: the model generates function calls with arguments and returns structured JSON. This allows it to be integrated into agent scenarios, request processing, or pipelines where the response needs to be parsed programmatically without manual text parsing.
The model accepts text and documents as input—their content can be parsed directly within the request. Images, audio, and video are not listed as supported input formats, so for such materials, you should choose another model from the BotHub catalog. Text and document tasks remain its specialty.
There are no specific benchmarks for Russian in the model's data, so we cannot promise the level of larger models. This is a compact 1.7B parameter model, and it is reasonable to test it on your own examples for complex texts. Trial requests in BotHub are pay-as-you-go, without a subscription.