LFM3-VL-2B vs Mistral Large 4

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.

Mistral Large 4 50

Memory scores 0, since the smallest 4-bit build, Mistral's NVFP4, is 610 GB, about 732 GB with a fifth added, far above 256 GB. Runtimes score 10, released Transformers and llama.cpp have the architecture, while vLLM v0.33.0 and SGLang v0.5.23 still lack it, the llama.cpp build Ollama 0.40.14 pins predates it, and LM Studio has no entry. Builds score 15 for Mistral's FP8 and NVFP4 releases. Terms score 10, a modified MIT licence that is free commercially except for Model-as-a-Service businesses. Download scores 10 with no gate, and hosted too scores 5 because Mistral's API serves these weights. 0 + 10 + 15 + 10 + 10 + 5 = 50.

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.

Mistral Large 4 75

File safety scores 25, safetensors only and loaded by Transformers' own code with a Jinja template. Provenance scores 15, Mistral's verified organisation with a hash on every file. Safety evaluation scores 10, refusal and prompt-injection results Mistral ran itself, listed in its 2 November technical report as internal, and red-teaming by outside firms with nothing published. Agent guidance scores 10 for a statement of what the refusals cover and a claim of trained injection resistance, with only general safeguards. Licence risk scores 10, 5 off for the Model-as-a-Service limit. The security programme scores 5 for security.txt with no public advisories. 25 + 15 + 10 + 10 + 10 + 5 = 75.

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.

Mistral Large 4 64

The card covers intended use, architecture, parameters, context and languages, and misses limitations and a cut-off (14). Training data is described by its mix in the technical report, with no sources, cut-off or filtering (5). The files changed on 2 November 2026, when the report was added (20). A Mistral member replied in three of nine discussions (10). It's a first release with older Large models in their own repositories and no dated note (5), and the files agree with each other and the card (10). 14 + 5 + 20 + 10 + 5 + 10 = 64.

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.

Mistral Large 4 95

The chat template is in the repository (20). The card documents how tools are declared and come back and names the parser (25), and the template renders several calls in one turn (20). Strict JSON Schema output is documented with how to ask for it (15). Reasoning effort is budgeted per request with the reasoning returned apart from the answer (10). Context scores 5, because the 1M window comes from YaRN scaling and the technical report gives 40K as the longest pretraining sequence. 20 + 25 + 20 + 15 + 10 + 5 = 95.