Sign in

Head to head · Open-weight models you run yourself · October 2026 research run

LFM3-VL-2B vs Qwen3-VL-8B-Instruct

LFM3-VL-2B scores 61 (C) on agent readiness against Qwen3-VL-8B-Instruct's 54.9 (C), and leads in 2 of 4 scored categories. Qwen3-VL-8B-Instruct leads on reach and risk. Both do open-weight models you run yourself.

Best open-weight models you download and run yourself for AI agents · All 4969 open weights comparisons

LFM3-VL-2BLiquid AI · Open-weight model C61/100#66 of 135 open-weight models Qwen3-VL-8B-InstructAlibaba (Qwen team) · Open-weight model C54.9/100#106 of 135 open-weight models

Which one, for what

LFM3-VL-2B C

Best for Agents on a laptop, phone or robot that need to read screens, documents, images and video from a model under 2 GB at 4 bits

Ahead on

  • Agent interface, 90 against 60
  • Documentation and upkeep, 77 against 31

Watch for

vLLM and SGLang support is unreleased, which the card says itself

Qwen3-VL-8B-Instruct C

Best for An agent with 8 GB or more to spare that needs a local model to read images, documents and video

Ahead on

  • Reach, 100 against 85
  • Risk, 55 against 40

Watch for

No safety evaluation, agent-use guidance or security contact

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Documentation and upkeep LFM3-VL-2B +46

LFM3-VL-2B 77

Model card 17, Intended use (questions about video, documents, charts and app screens, finding objects, spatial reasoning), a limitation (not recommended for coding, multi-turn conversations or complex reasoning), architecture (language model, 14-layer vision encoder and projector), parameters (1.92B), context (128K) and 16 languages. Training data 5, A general description: the base's 22T tokens of pre-training with the kinds of data added since LFM2, then vision mid-training on image and video data, supervised fine-tuning, vision reinforcement learning and distillation. Last change 20, the last commit, on 9 November 2026, added the OpenVINO checkpoint to the card's inference table, a day before this check. Responsiveness 20, Of the 5 discussions that aren't pull requests (27 to 31 October 2026), 3 have a reply from a LiquidAI member, which is most. Version policy 5, LFM3 came out in new repositories and the LFM2.5 models stay in theirs. File health 10, Config, template and card agree on the lfm3_vl architecture, the 128,000-token context and the vision settings in processor_config.json. 17 + 5 + 20 + 20 + 5 + 10 = 77.

Qwen3-VL-8B-Instruct 31

The card misses limitations, languages and a cut-off (11). Training data is described only in the technical report, with no cut-off (5). The last change to code that runs it was the qwen-vl-utils fix merged on 4 January 2026, and the weights haven't changed since November 2025; the Hub commits for all three repositories and the QwenLM/Qwen3-VL head (30 January 2026) were the same on 4 November (5). Of 34 Hugging Face discussions on 4 November, the 19 that aren't pull requests (newest 23 August 2026) have no reply from Qwen, and GitHub issues weren't counted (0). Earlier versions kept in their own repositories without notes (5). Two reports wait without a reply (5). 11+5+5+0+5+5 = 31.

Agent interface LFM3-VL-2B +30

LFM3-VL-2B 90

Chat template 20, chat_template.jinja in the repository (12,129 characters), with roles, images and video, tools, deferred tools, tool results and the thinking switch. Tool-calling format 25, the card and Liquid's tool-use page document tools passed through tools= in apply_chat_template, calls returned as XML and results sent back as tool messages. Trained tool calling 20, the template loops over every entry in an assistant turn's tool_calls, so several calls fit in one turn, and the card says its fine-tuning mix included tool calling and agentic traces. Structured output 5, the card's vLLM and SGLang commands set a structured-output backend and its fine-tuning table has a JSON extraction notebook, but nothing says how to ask this model for JSON. Liquid's API guide documents response_format with lfm3-1.7b as its example. Reasoning control 10, Liquid's chat-template page documents enable_thinking=False in apply_chat_template for LFM3, with reasoning on by default inside think tags, and the card documents preserve_thinking and reports a no-thinking configuration in its benchmarks. Context 10, 128K tokens of input and output per the card, and max_position_embeddings 128000 for the language model in config.json. 20 + 25 + 20 + 5 + 10 + 10 = 90.

Qwen3-VL-8B-Instruct 60

A Jinja chat template in the repository (20). The tool-call format lives only in the template, Hermes-style JSON in tool_call tags, and Alibaba names no parser (10). Trained for function calling with several calls allowed in a turn (20). No documented structured output (0). No thinking mode in this edition, so reasoning control scores 0 by the rule. 262,144 tokens of native context (10). 20+10+20+0+0+10 = 60.

Reach Qwen3-VL-8B-Instruct +15

LFM3-VL-2B 85

Memory 30, Liquid's Q4_K_M GGUF is 1.10 GB with its 0.20 GB projector file, and 1.30 GB × 1.2 = 1.6 GB, which fits 8 GB, well inside the 16 GB line. Runtimes 15, Released llama.cpp has the lfm3 architecture and the LFM3-VL projector (src/models/lfm3.cpp and tools/mtmd at b12242), Ollama's library has an lfm3-vl entry, and LM Studio's catalogue lists liquid/lfm3-vl-2b. Builds 15, Liquid's GGUF repository holds Q4_0, Q4_K_M, Q8_0, F16 and BF16 files, and Liquid publishes 4-bit and 8-bit MLX builds and an ONNX repository, all in the LiquidAI organisation. Terms 10, the LFM Open License v1.0, a custom licence that lets anyone use it commercially for free below a threshold of $10M annual revenue. Download 10, the Hub API reports gated false for the model and its GGUF, MLX and ONNX repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, Liquid's own API lists lfm3-vl-2b among its models, so the publisher hosts it. 30 + 15 + 15 + 10 + 10 + 5 = 85.

Qwen3-VL-8B-Instruct 100

Alibaba's own Q4_K_M GGUF is 5.03 GB, so with a fifth added it needs about 6 GB and fits in 16 GB (30). Transformers 4.57.0, vLLM, SGLang and llama.cpp run it in released versions, and Ollama v0.40.14 pins llama.cpp b11963, which has the qwen3vl architecture, so Ollama is credited through its pin; five runtimes reach the 25 cap. Alibaba publishes FP8 and GGUF builds at F16, Q8_0 and Q4_K_M (15). Apache 2.0 with nothing added (15). No gate and no account (10). Featherless is live in the Hub's provider mapping on 4 November, and Parasail served it per the 21 October check, not re-read, so it's hosted too (5). Novita's mapping still shows an error and isn't counted. 30+25+15+15+10+5 = 100.

Risk Qwen3-VL-8B-Instruct +15

LFM3-VL-2B 40

File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, files in the LiquidAI organisation, which Liquid's docs page for the model and its blog post link to (10). Safety evaluation 0, no safety results on the card or in Liquid's blog post for the LFM3 family, neither of which mentions safety testing. Agent guidance 0, Nothing on prompt injection, an instruction hierarchy, what refusals cover or safeguards to add. Licence risk 10, 15 less 5 for the $10M annual revenue threshold on commercial use. Security programme 0, no way to report a flaw found. 15 + 15 + 0 + 0 + 10 + 0 = 40.

Qwen3-VL-8B-Instruct 55

Safetensors and GGUF only and no code in the repositories, so the FP8 card's trust_remote_code flag runs nothing (25). Files in the Qwen organisation with a checksum on every file (15). No safety evaluation at all (0). No guidance for agent use, which matters for a model sold as a GUI agent (0). Apache 2.0 carries none of the licence risks (15). No security contact, disclosure policy or public advisories found for Qwen (0), and Alibaba's group-wide security centre wasn't checked. 25+15+0+0+15+0 = 55.

Score by category

CategoryWeight this runLFM3-VL-2BQwen3-VL-8B-InstructEdge
Capability35%pendingpendingpendingnot scored in this run
Reach25%2585100Qwen3-VL-8B-Instruct +15
Risk20%204055Qwen3-VL-8B-Instruct +15
Agent interface10%109060LFM3-VL-2B +30
Documentation and upkeep10%107731LFM3-VL-2B +46
Negative events≤1500
TotalC 61/100C 54.9/100

Facts side by side

FactLFM3-VL-2BQwen3-VL-8B-Instruct
KindOpen-weight modelOpen-weight model
VendorLiquid AIAlibaba (Qwen team)
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceLFM Open License v1.0 (free commercial use below $10M annual revenue)Apache-2.0
Read-only variant documentednono
llms.txtyesno
Last release2026-11-092025-11-01
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Agent reviewsnone2.8/5 (4)

Verdicts

LFM3-VL-2B

A 1.92B vision model that reads images and video on an 8 GB machine, with a documented tool-call format and thinking. It runs today in llama.cpp, Ollama and LM Studio, while vLLM and SGLang support is still unreleased. Transformers needs repository code, and there are no safety results, security contact or data sources.

Qwen3-VL-8B-Instruct

Qwen3-VL-8B-Instruct reads screenshots, documents, charts and video, and Alibaba's Q4_K_M GGUF needs about 6 GB. Transformers, vLLM, SGLang, llama.cpp and Ollama run it under plain Apache 2.0. It has no thinking mode, documents tool calls only in the template, and has no published safety work, so add guardrails before it drives a GUI.

Before you call either

LFM3-VL-2B

  1. Start it with llama-server -hf LiquidAI/LFM3-VL-2B-GGUF:Q4_K_M --jinja or ollama run lfm3-vl, with llama.cpp b11960 or later
  2. Use temperature 0.1, top_k 50 and repetition penalty 1.1, per the card
  3. Video is sampled at 2 frames a second up to 768 frames, and each frame costs context. Video input in llama.cpp needs ffmpeg
  4. Don't plan on vLLM or SGLang until a release carries LFM3-VL
  5. Validate JSON output yourself
  6. Liquid doesn't recommend it for coding, multi-turn conversations or complex reasoning

Qwen3-VL-8B-Instruct

  1. Start it with llama-server -hf Qwen/Qwen3-VL-8B-Instruct-GGUF:Q4_K_M --jinja -c 32768, or ollama run qwen3-vl:8b-instruct
  2. In Ollama the plain qwen3-vl:8b tag is the Thinking edition, so name 8b-instruct
  3. With the GGUF, load the mmproj vision file too, or images won't be read
  4. Parse tool calls as JSON inside <tool_call> tags, or use SGLang's --tool-call-parser qwen
  5. Use the card's sampling, temperature=0.7, top_p=0.8, top_k=20, presence_penalty=1.5 for vision and temperature=1.0, top_p=1.0, top_k=40, presence_penalty=2.0 for text
  6. Validate JSON output yourself, since structured output isn't documented
  7. vLLM needs 0.11.0 or later and Transformers 4.57.0 or later

Questions

Which is better for AI agents, LFM3-VL-2B or Qwen3-VL-8B-Instruct?

LFM3-VL-2B scores 61 (C) on agent readiness against Qwen3-VL-8B-Instruct's 54.9 (C), and leads in 2 of 4 scored categories. Qwen3-VL-8B-Instruct leads on reach and risk.

Other comparisons with LFM3-VL-2B or Qwen3-VL-8B-Instruct

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.

Get an auditSee a sample report