Moondream 3 vs LFM2.5-VL 3B : l'un pointe la réponse, l'autre se contente de répondre

The most useful difference between M​oondream 3 Preview and LFM2.5-VL-3B is not in their benchmark tables. It is that M​oondream 3 Preview reasons about an image by pointing into it — its reasoning traces are grounded, referencing specific regions — while LFM2.5-VL-3B is explicitly a non-reasoning model that answers directly, a choice Liquid AI made to keep latency low for on-device work. Same image in, two entirely different things come out, and which one you want depends on whether a downstream system has to trust the answer or merely consume it.

Everything else follows from that fork. M​oondream 3 Preview is a sparse mixture-of-experts model at 9B total parameters with 2B active, released on 18 September 2025 by M​oondream, which was acquired by Prime Intellect in October 2026. LFM2.5-VL-3B is a dense 3.1B model — a 2.6B language backbone with a 400M SigLIP2 NaFlex vision encoder — released by Liquid AI on 12 August 2026 as a direct successor to LFM2-VL-3B. One is a specialist that was built to detect and point and can also answer questions. The other is a generalist that was built to answer and recently learned to point.

La collision de métriques que vous devez repérer en premier

Mettez les deux chiffres de référence côte à côte : la comparaison ressemble à une quasi-égalité. Ce n’est pas le cas, car il ne s’agit pas de la même mesure.

M​oondream's launch chart lists M​oondream 3 Preview at 91.1 on RefCOCO. Liquid AI reports LFM2.5-VL-3B at RefCOCO precision@1 of 87.9. RefCOCO accuracy and RefCOCO precision@1 are different statistics computed over the same dataset, and a 3.2-point gap between them tells you nothing about which model localises better. Anyone stacking those two figures in a bar chart is comparing a number to a differently-defined number.

• M​oondream 3 Preview grounding — RefCOCO 91.1, on a chart where frontier baselines (GPT-5, Gemini 2.5-Flash, C​laude 4 S​onnet) were queried through a template because they have no native grounding, and where some preview scores came from a 100-question sample rather than the full set. Vendor-reported and, as M​oondream says itself, not a fair fight.

• Ancrage LFM2.5-VL-3B — précision@1 sur RefCOCO de 87,9, contre 57,1 pour le précédent LFM2-VL-3B, ce que Liquid attribue à la mise à l’échelle des données d’ancrage synthétiques. Rapporté par le fournisseur, comparaison à harnais identique face à son propre prédécesseur, ce qui est l’axe le plus pertinent.

• Detection and pointing — M​oondream 3 Preview: native detect and point skills, structured coordinate output, part of the architecture. LFM2.5-VL-3B: grounding added by training, returned as coordinates the same way, with function calling also new to the VL line (ToolSandbox 26.4 to 59.5, BFCL v4 20.5 to 32.5).

The honest read: Liquid AI has a clean before-and-after on identical harnesses and M​oondream has a cross-model chart with template caveats. Neither gives you a head-to-head. If grounding accuracy is the deciding factor in your evaluation, your own labelled set is the only instrument that will settle it.

Ce que vous coûte un raisonnement ancré

M​oondream's query skill runs in reasoning mode by default. The model works through the question and its trace references the regions of the image it is using, which you can inspect in the playground by hovering over the text. For anything where a human has to audit the answer — a defect call on a production line, a security incident, a medical triage flag — that property is worth a lot. It is the difference between a model that says "no" and a model that says "no, here." p>The cost is tokens. Reasoning mode generates a trace before the answer, and M​oondream's own documentation offers the escape hatch: pass reasoning=False for simple questions, because the company knows the mode is not free. That is a real architectural admission — a well-designed model would not need a flag to switch the thought off.

LFM2.5-VL-3B n’a pas d’équivalent. Il a été conçu comme un modèle sans raisonnement, et Liquid AI assume clairement ce compromis : répondre directement permet de garder une faible latence pour les usages en temps réel et sur appareil. Il n’y a aucune trace à inspecter et aucun mode à désactiver. À une caisse en train de lire des codes-barres, ou sur un téléphone qui traduit un menu, c’est le bon choix d’ingénierie, et l’absence de trace n’est pas un défaut.

So the question is not which is better. It is whether anything downstream of your model reads the reasoning. If a human does, M​oondream's grounded traces are the feature you are buying. If only a program does, you are paying for tokens nobody will read.

Là où 3.1B gagne réellement

L'argument de taille de LFM2.5-VL-3B n'est pas du marketing. Il tourne en moins de 3,3 Go de mémoire à 228 tokens par seconde sur un Apple M5 Max, et à 116 tokens par seconde sur un AMD Ryzen AI Max+ 395, avec des exports GGUF, ONNX et MLX publiés en parallèle du checkpoint natif. C'est un modèle que vous pouvez installer sur un ordinateur portable, et Liquid AI le propose aussi sous forme de Space WebGPU Hugging Face qui s'exécute dans un onglet de navigateur.

M​oondream 3 Preview has a smaller active parameter count — 2B against 3.1B — but the comparison is not symmetric, because a MoE checkpoint has to be fully resident even though only a fraction of it fires. The 9B of weights need somewhere to live. M​oondream's own answer to that is Photon, its local inference engine, which runs the model on N​VIDIA GPUs from Ampere onward and on Apple Silicon Macs. Different constraint: LFM2.5-VL-3B is small enough to fit anywhere, M​oondream 3 Preview is fast in the compute sense but large in the memory sense.

• Context — both 32K tokens. Genuine convergence, and worth noting given the two-year jump from the 2K windows of the previous M​oondream generation.

• Languages — LFM2.5-VL-3B lists sixteen: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, Polish, and doubled its vocabulary to 128,000 to support non-Latin scripts. M​oondream 3 Preview's card leads on visual capability rather than a language list; treat multilingual VQA as something to test rather than something promised.

• Document reading — LFM2.5-VL-3B: DocVQA 91.1, ChartQA 81.3, OCRBench v1 84.2, OCRBench v2 English 47.5, MMMU 48.4. M​oondream 3 Preview: DocVQA 88.3, ChartQA 86.6, POPE 89.0. Different harnesses, vendor-reported on both sides.

• Ce que LFM2.5-VL-3B vous dit de ne pas faire — la fiche de Liquid AI indique clairement qu'il n'est pas recommandé pour les tâches à contexte long ou à forte intensité de raisonnement, comme la conception web visuelle ou les questions techniques sur des plans. Cette phrase est plus utile que la plupart des tableaux de benchmarks.

Déploiement, et où un routeur intervient

Neither of these models is served through OrcaRouter, and it is worth saying so directly rather than dancing around it: both are open-weight checkpoints you run yourself, and M​oondream 3 Preview's hosted endpoint — M​oondream Cloud — is being shut down on 13 November 2026 following the Prime Intellect acquisition, with unspent credits refunded.

Ce que cela change, c’est la forme de la décision. Une équipe qui standardise sur un VLM auto-hébergé finit généralement par ajouter un second modèle, hébergé, pour les cas que le modèle local gère mal, et à ce moment-là, le travail d’intégration devient le problème : deux SDK, deux mécanismes d’authentification, deux politiques de réessai, et une panne de fournisseur qui devient votre panne.

This is the case OrcaRouter's routing DSL is built for. You compose several models into a single call — the 3.1B local model answers what it can, a hosted model takes the residue — behind one key for more than 200 models, with provider list price passed through at zero markup and automatic failover when an upstream degrades. Since neither M​oondream 3 Preview nor LFM2.5-VL-3B is in that catalogue, the routing layer sits above them rather than replacing them: it is how you keep the cheap local path cheap without making a vendor outage your outage.

Le fork des licences, plus tranché que les benchmarks

Deux philosophies de licence différentes, et la différence compte davantage pour une décision produit qu’un écart de score de 3 points.

M​oondream 3 Preview ships under the Business Source License 1.1 with an Additional Use Grant (No Third-Party Service). Internal and production use, personal projects, research, benchmarks, fine-tunes, merges, quantisations, weight deltas, and use inside a product that does not substantially overlap M​oondream's paid offerings are all permitted. Selling a hosted API that exposes similar vision capabilities, managed "M​oondream-as-a-service", or embedding the weights in a paid SDK that delivers comparable capabilities requires a separate agreement. M​oondream 3.1 moved to the M​oondream Model License 1.0, which states the same rule more cleanly: component of your product is allowed, access to M​oondream itself as the product is not. Edge devices, customer-controlled VPCs, and running the model on leased cloud or neocloud GPUs are all explicitly on the allowed side.

LFM2.5-VL-3B is under Liquid AI's own LFM Open License v1.0. Read it before you ship, as you would any bespoke licence, but the families are different in kind: M​oondream's restriction targets one specific business — reselling the model — while Liquid's is a broad open licence with its own terms.

If you are a domain-specific builder, both are fine and the licence question is a formality. If your business is anywhere near "we host vision models for other people," M​oondream's licence is the constraint you read first and LFM's is the one you probably clear.

Lequel choisir

Choose M​oondream 3 Preview if the answer has to be inspectable. If a human reviews the output, if the trace needs to show which part of the image drove the decision, if detection and pointing are the primary product rather than a feature — that grounded reasoning is the whole reason to pay 9B of memory for a 2B-active model. Budget for Photon or a GPU, and plan around the hosted endpoint closing.

Choisissez LFM2.5-VL-3B si le modèle doit résider sur l'appareil. Si vous déployez vers un téléphone, un ordinateur portable, un boîtier d'atelier ou un onglet de navigateur, et que la charge de travail est à tour unique, à haut débit et à faible latence — lire des écrans, transcrire des documents, détecter des objets dans un flux — alors un modèle dense de 3,1B qui répond sans réfléchir est exactement ce qu'il vous faut, et ses 228 tokens par seconde sur un M5 Max est le chiffre qui tranche.

Pick both if you have the budget for it, and route between them: LFM2.5-VL-3B on every frame because it is nearly free to run, M​oondream 3 Preview on the frames where a grounded answer is worth waiting for.

Ce que les chiffres ne peuvent pas vous dire, c'est si quelque chose a mal tourné sur le terrain : les incidents d'IA documentés et les défaillances d'agents concernant les modèles de cette classe sont recensés dans l'OrcaRouter AI Incident Archive, où chaque entrée comporte ses propres sources, sa gravité, son niveau de confiance et son indicateur de contestation.