Key takeaways
- OpenRouter added liquid/lfm-3-1.7b and liquid/lfm-3-1.7b:free at 23:57:57 UTC on November 6, 2026. Both point to Liquid AI’s 1.73B-parameter, text-only LFM3-1.7B checkpoint.
- The free route lists $0 input and output pricing. The paid route lists $0.08 per million input tokens, $0.02 per million cached-input tokens and $0.20 per million output tokens.
- Both registry entries expose a 262,144-token context and an 8,192-token maximum completion. OpenRouter marks reasoning as mandatory and lists chat completions, completions, Messages and Responses API surfaces.
- OpenRouter showed one provider, Liquid, and warned that prompts and outputs may be retained and used to train Liquid models. Do not send personal, customer, proprietary or regulated data merely because the route is free.
- Liquid AI positions the model for local agents, tool calling, structured output and specialization, but its own card says the small model is not recommended for agentic coding or complex multi-step reasoning.
- Use the free route for non-sensitive evaluation, then compare its reliability and total operating fit with the downloadable weights or paid route. AccessAllGPT could not run a hands-on test because no OpenRouter key was available in this job.
OpenRouter added two LFM3-1.7B routes on November 6
OpenRouter registered Liquid AI’s LFM3-1.7B at 23:57:57 UTC on November 6, 2026. Builders can now request liquid/lfm-3-1.7b as a paid route or liquid/lfm-3-1.7b:free as a zero-priced trial route. Both entries resolve to the same canonical model record and name LiquidAI/LFM3-1.7B as the source checkpoint.
This is a new managed access route, not a new underlying checkpoint. Liquid AI launched LFM3 on October 27 with downloadable weights, a Playground and its own API. The OpenRouter addition matters because it puts the model behind APIs and request formats that many multi-model applications already use, while adding a free option for evaluation.
The free route costs zero; the paid route starts at $0.08 per million input tokens
OpenRouter lists zero prompt and completion prices for liquid/lfm-3-1.7b:free. For liquid/lfm-3-1.7b, the public registry lists $0.08 per million input tokens, $0.02 per million cached-input tokens and $0.20 per million output tokens. Those are registry prices retrieved on November 7 and can change; verify the route and invoice before committing production traffic.
Free does not mean contractually durable. When reviewed, the model page showed one provider rather than a redundant pool. Teams should test quota behavior, queueing, regional availability, error handling and withdrawal risk. The paid ID may be easier to budget, but the public record alone does not establish a service-level agreement.
The API exposes 262K context but caps completions at 8,192 tokens
Both entries advertise a 262,144-token combined context and text-only input and output. The top-provider record caps one completion at 8,192 tokens. A large context label therefore does not mean one request can generate hundreds of thousands of tokens, nor does it establish accurate retrieval at every position.
OpenRouter marks reasoning as mandatory and lists tool calls, tool choice, structured output, response formats, seeds, log probabilities and common sampling controls. Its defaults are temperature 0.1, top-k 50 and repetition penalty 1.1. The registry exposes chat completions, legacy completions, Anthropic-compatible Messages and Responses API surfaces; compatibility still needs case-level testing.
The data policy is the main catch
The model page says prompts and outputs may be retained and used to train Liquid models, and the only displayed provider carried OpenRouter’s “Trains” privacy label. Treat that as a hard data-boundary fact, not a footnote. A zero token price does not offset the risk of sending source code, unreleased product plans, customer records, credentials or regulated information into a training-enabled route.
For early trials, use synthetic or already-public prompts and strip secrets from traces. Before production, obtain the current provider policy, retention period, deletion process, regional path, subprocessors and contractual controls. If the workload cannot tolerate training use, choose a route whose data policy meets that requirement or run the downloadable checkpoint under an approved local setup.
Liquid AI says the model is for bounded agent work, not hard coding problems
Liquid AI’s card describes LFM3-1.7B as a 1.73B-parameter, 24-layer text model for local agents, tool calling, structured outputs and domain-specific fine-tuning. It supports 16 listed languages and uses the LFM Open License v1.0 for downloadable weights. The company says a 4-bit build takes about 1 GB of memory; AccessAllGPT did not reproduce that sizing claim.
The developer also supplies an unusually direct limitation: use retrieval or search for knowledge-heavy tasks, and do not treat this small model as the default for agentic coding or complex multi-step reasoning. That makes the OpenRouter route best suited to narrow extraction, formatting, classification and bounded tool-selection trials rather than unsupervised software changes or long plans.
What builders should do next
Start with the free ID on a frozen set of non-sensitive cases. Test exact JSON schemas, tool arguments, refusal and error behavior, long-context retrieval at several positions, latency under realistic concurrency and recovery after malformed outputs. Compare every result with a simple baseline and log the exact model ID, route, date and request settings.
Move to the paid route only if it improves availability or governance enough to justify the dependency. Consider local weights when data control and offline operation matter, but review the custom license and serving stack first. Wait when the provider’s training use is incompatible with the workload; reject the model when representative quality, completion length, latency or tool reliability misses the application’s threshold.
Copy-ready LFM3 OpenRouter trial record
Complete this for the free and paid IDs separately so zero price does not hide differences in data treatment, reliability or operating fit.
Entries stay in this browser tab and are not submitted to AccessAllGPT. Blank responses are copied as [Unresolved].
Exact model ID, retrieval date, free or paid tier, provider shown, API surface, region and fallback.
Allowed data classes, training use, retention and deletion terms, secrets filtering, subprocessors and approval owner.
System prompt, context size, maximum completion, reasoning setting, tools, response schema and sampling values.
Representative and adversarial cases, long-context positions, expected structured outputs, baseline and acceptance thresholds.
Success rate, malformed output, tool errors, refusals, latency, throughput, quota errors and retry cost.
Current token rates, cache rate, free-tier limits, monthly estimate, model-withdrawal plan and paid or local fallback.
Trial, constrain, approve, wait or reject; allowed workload, owner, expiry and revalidation trigger.
Primary sources
Browse the publication-wide evidence index →
- LFM3-1.7B (free) model pageOpenRouter · Reviewed: Model identity; release date; description; free price; context; provider count; provider privacy label; live latency, throughput and uptime panels · Retrieved · Supports: OpenRouter lists liquid/lfm-3-1.7b:free as a free text model released November 6, 2026, with a 262,144-token context. Its only listed provider was Liquid and carried a “Trains” privacy label when reviewed.
- OpenRouter Models API entries for LFM3-1.7BOpenRouter · Reviewed: IDs; canonical slug; creation timestamp; architecture; pricing; context and completion limits; supported parameters; defaults; API endpoints; reasoning requirement · Retrieved · Supports: The registry exposes paid and free IDs created at 23:57:57 UTC on November 6. Both are text-to-text routes with a 262,144-token context and 8,192-token maximum completion. The paid route lists $0.08/M input, $0.02/M cached input and $0.20/M output; the free route lists $0 prompt and completion prices.
- LFM3-1.7B model cardLiquid AI on Hugging Face · Reviewed: Model summary; parameter count; architecture; context; languages; license; intended uses and limitations; inference variants · Retrieved · Supports: Liquid AI describes LFM3-1.7B as a 1.73B-parameter post-trained text model for local agents, tool calling, structured output and specialization. It warns that the small model is not recommended for agentic coding or complex multi-step reasoning.
- Introducing LFM3: Every Modality for Specialized IntelligenceLiquid AI · Reviewed: October 27 launch date; family availability; architecture and training description; context and edge claims; benchmark sections · Retrieved · Supports: Liquid AI announced the LFM3 family on October 27 and made weights available through Hugging Face, its Playground and its own API. Architecture, speed, memory and quality results in the announcement are vendor claims.
Limitations
This is a source-backed API launch briefing, not hands-on model testing. No OPENROUTER_API_KEY was available in the publishing environment, so AccessAllGPT did not call liquid/lfm-3-1.7b or liquid/lfm-3-1.7b:free; inspect outputs or hidden reasoning; test tools, schemas, API compatibility, multilingual behavior, long-context retrieval or completion limits; measure latency, throughput, uptime, queueing or rate limits; compare free and paid routing; download or serve the weights; reproduce Liquid AI benchmarks or memory claims; audit training data; or obtain production terms. OpenRouter’s live performance panels, provider availability, prices, privacy labels and model records can change. Liquid AI’s architecture, training, quality, speed and memory statements are vendor claims.
Disclosures
AccessAllGPT received no OpenRouter or Liquid AI credits, API key, model access, weights, hardware, briefing, review, payment or compensation for this article. OpenRouter, Liquid AI and Hugging Face did not sponsor, review or endorse it. AccessAllGPT did not test, score or rank LFM3-1.7B. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with OpenRouter, Liquid AI or Hugging Face. Publication-wide relationships are listed on the disclosures page.
