The best vision and multimodal AI models in 2026
The best vision and multimodal AI models right now are SAM 3 (Segment Anything Model 3), Moondream 3.1 (9B-A2B) and LongCat-2.5, out of 19 we have reviewed. Each is ranked by our rating, with its price, our verdict and who it suits.
By ReviewRamble editors · Updated Nov 15, 2026 · How we review
How we ranked: by our rating out of 5, from reviews built on each product's own pricing, documentation and policies plus independent benchmarks where they exist, each checked on the date shown in the review. We take no affiliate commission.
- 1
SAM 3 turns "outline every forklift in this video" into one prompt, at more than twice the accuracy of the open-vocabulary detectors before it, though still well short of a human labeller.
Best for: Computer-vision teams that need masks or boxes for named objects across images and video without training a detector for each class.
Price: Free under the SAM License after a gated sign-up on Hugging Face; commercial use allowed with military, ITAR and reverse-engineering restrictions. Hosted use through Roboflow is billed on Roboflow's plans.
Read our SAM 3 (Segment Anything Model 3) review4.4/5 - 2
For finding and counting things in images at volume, Moondream 3.1 is faster and more accurate than the general vision models people usually reach for.
Best for: Engineering teams that need open-vocabulary detection, pointing or counting on many images a second, on their own GPUs or a cheap API.
Price: Free weights under the Moondream Model License. Moondream Cloud and Cloudflare Workers AI charge $0.30 per million input tokens and $1.00 per million output tokens, with 50% off through the Batch API.
Read our Moondream 3.1 (9B-A2B) review4.2/5 - 3
LongCat-2.5 is among the strongest open agentic models on paper, under a licence with almost no strings, but at 1.6 trillion parameters it is a model for clusters, not companies with one GPU server.
Best for: Labs and large companies with serious inference hardware that want an open, permissively licensed agentic model to host themselves.
Price: Free open weights under the MIT licence; running it needs a large multi-accelerator cluster.
Read our LongCat-2.5 review4.1/5 - 4
This is the open model to start from if you want a local assistant that sees and talks in real time, as long as English or Chinese is enough.
Best for: Developers building a private, local voice-and-camera assistant in English or Chinese on one GPU or a high-end Mac.
Price: Free Apache 2.0 weights. No first-party hosted price published; you pay for your own hardware.
Read our MiniCPM-o 4.5 review4.1/5 - 5
The open model to beat at 120B on independent tests, provided the job is documents and research rather than code.
Best for: Teams running document and research agents who want the strongest open mid-size model, especially on NVIDIA hardware.
Price: Free weights under OpenMDW-1.1. Hosted: OpenRouter from $0.15 input / $0.50 output per 1M tokens, plus a free variant; NVIDIA NIM and other providers priced separately.
Read our Nemotron 3.5 Super VL review4.1/5 - 6
Sapiens2 is the most detailed open model for reading human bodies in a photo, under a licence strict enough that many of the obvious products need a lawyer first.
Best for: Graphics, AR and fitness teams that need dense, high-resolution pose, body-part masks or 3D shape for people in photos and video frames.
Price: Free under the Sapiens2 License; no sign-up gate. Bans surveillance, biometric processing, re-identification, deepfakes, military and ITAR-related uses, and allows Meta to audit compliance. No hosted API.
Read our Sapiens2 review4.1/5 - 7
On OpenBMB's own numbers this is the open model to beat for reading documents and charts at the 35B size, and the 1B is the rare phone-sized vision model worth shipping.
Best for: Teams extracting text and tables from scanned documents and charts on their own hardware, and mobile developers who need a vision model under 2 GB.
Price: Free Apache 2.0 weights for both sizes. No first-party hosted price published; you pay for your own hardware.
Read our MiniCPM-V 4.7 (35B-A3B and 1B) review4.0/5 - 8
A cheap, capable model for vision and office-style agent work that you cannot yet download, which is half of what the Inkling name promised.
Best for: Teams that want a low-cost model for screen reading, document vision and finance or office agents, and plan to fine-tune it on Tinker.
Price: Tinker: $0.30 per 1M input tokens ($0.03 cached), $1.00 per 1M output. Training $1.50 per 1M tokens at 64K context, $3.00 at 256K. Serverless inference is in beta.
Read our Inkling 2 Small review3.9/5 - 9
Mage-VL-Cowork gets a 4B model within a few points of the best open desktop agents by letting it reach for a terminal, but Microsoft labels it research-only and Linux is all it has been tested on.
Best for: Researchers and agent-platform teams studying small computer-use agents on Linux, who can run them in disposable sandboxes.
Price: Free under Apache 2.0, though Microsoft's repository says the Mage models are for research and not intended for product or service deployment. No hosted API.
Read our Mage-VL-Cowork review3.9/5 - 10
MedGemma 1.5 is the open starting point for medical imaging apps that need 3D scans, but it is a base to validate and adapt, not a model to put in front of patients.
Best for: Health-tech and hospital engineering teams building and validating their own imaging or records tools on premises.
Price: Free weights under the Health AI Developer Foundations terms. Vertex AI hosting is billed at Google Cloud compute rates.
Read our MedGemma 1.5 4B review3.9/5 - 11
Molmo2-8B is the open model to reach for when you need to know where in a video something is, not just whether it is there.
Best for: Researchers and robotics or video-analytics teams that need open, explainable video grounding they can retrain.
Price: Free weights under Apache 2.0; some training datasets are non-commercial. No paid hosted API listed.
Read our Molmo2-8B review3.9/5 - 12
DiffusionGemma makes a capable open model answer one user several times faster on one GPU, and Google itself says to stick with plain Gemma 4 when quality matters.
Best for: Developers building single-user, latency-sensitive tools such as inline editors or autocomplete on one NVIDIA GPU.
Price: Free Apache 2.0 weights. No first-party per-token price published; you pay for your own GPU or cloud deployment.
Read our DiffusionGemma 26B A4B review3.8/5 - 13
Inkling is the large open model you pick for its licence and its ears, not for its eyes.
Best for: Teams that want a permissively licensed open model reading text, images and audio, and plan to fine-tune it rather than use it as is.
Price: Free weights under Apache 2.0. OpenRouter (Together): $1.00 per million input tokens, $4.05 per million output, 524,288-token context. Tinker prices per its docs.
Read our Inkling review3.8/5 - 14
LocateAnything-3B finds things in pictures faster than the open models NVIDIA tested it against, and NVIDIA will not let you ship it.
Best for: University and non-profit labs that need fast open-vocabulary detection or GUI grounding for research.
Price: Weights free for academic and non-profit research; commercial use not permitted. No hosted API.
Read our LocateAnything-3B review3.8/5 - 15
A specialist dressed as a generalist: on its makers' own numbers it reads rooms and robot scenes far better than Qwen3.5-9B, and reads text slightly worse.
Best for: Robotics and embodied-AI researchers who need a small open model that reasons about viewpoints, layouts and affordances, and can live with a forked serving stack.
Price: Free weights under the NVIDIA Open Model License, with the Qwen3.5 parts under Apache 2.0. No hosted API or per-token price is published; you pay for your own GPU.
Read our ZDTaichu5.0-9B review3.8/5 - 16
A sensible fine-tune of Qwen3.5-9B with real effort on Indian languages and on-device builds, but it has to prove it beats the model it started from.
Best for: Developers building offline or Indic-language apps who want a small multimodal model and will test it against Qwen3.5-9B themselves.
Price: Free open weights under Apache 2.0.
Read our Alpie Seed 9B review3.7/5 - 17
BAGEL is still the most useful open model for doing understanding, generation and editing in one place, but it is 18 months old and shows it.
Best for: Researchers and prototyping teams who need one open model that both reads and generates images and are willing to fine-tune it.
Price: Free Apache 2.0 weights and code. No first-party hosted API; a free public demo is available.
Read our BAGEL-7B-MoT review3.6/5 - 18
SenseNova-Loop-9B-Preview posts agent scores well above its size, but it is a preview with no technical report, a 38 GB download and serving support still in unmerged branches.
Best for: Research teams studying looped and weight-shared architectures, or evaluating small open models for office-agent work, who can tolerate preview-grade tooling.
Price: Free under Apache 2.0. No hosted API listed for this model.
Read our SenseNova-Loop-9B-Preview review3.5/5 - 19
Still the only Phi model that hears speech, but at 20 months old it has been overtaken by newer small multimodal models.
Best for: Developers who want a small MIT-licensed model for on-device speech and image input.
Price: Free weights under MIT. Azure AI Foundry: $0.08 input / $0.32 output per 1M text or image tokens; audio input $4 per 1M.
Read our Phi-4-multimodal review3.3/5
Sources
The main official source behind each ranked review; every review lists its full set.
- SAM 3 (Segment Anything Model 3): Hugging Face: facebook/sam3 (gated model page)
- Moondream 3.1 (9B-A2B): Hugging Face: Moondream 3.1 9B-A2B model card
- LongCat-2.5: meituan-longcat/LongCat-2.5 (Hugging Face)
- MiniCPM-o 4.5: Hugging Face: MiniCPM-o 4.5 model card
- Nemotron 3.5 Super VL: Hugging Face: NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 model card
- Sapiens2: Hugging Face: facebook/sapiens2 (model index)
- MiniCPM-V 4.7 (35B-A3B and 1B): Hugging Face: MiniCPM-V-4.7-35B-A3B model card
- Inkling 2 Small: Thinking Machines Lab: Inkling 2 Small model card
- Mage-VL-Cowork: Hugging Face: microsoft/Mage-VL-Cowork model card
- MedGemma 1.5 4B: Google: MedGemma model card
- Molmo2-8B: Hugging Face: allenai/Molmo2-8B model card
- DiffusionGemma 26B A4B: Hugging Face: DiffusionGemma 26B A4B model card
- Inkling: Hugging Face: thinkingmachines/Inkling model card
- LocateAnything-3B: Hugging Face: nvidia/LocateAnything-3B model card
- ZDTaichu5.0-9B: Hugging Face: TaichuAI/ZDTaichu5.0-9B model card
- Alpie Seed 9B: 169Pi/Alpie_seed (Hugging Face)
- BAGEL-7B-MoT: Hugging Face: BAGEL-7B-MoT model card
- SenseNova-Loop-9B-Preview: Hugging Face: sensenova/SenseNova-Loop-9B-Preview model card
- Phi-4-multimodal: Hugging Face: microsoft/Phi-4-multimodal-instruct model card
More guides
- Best large language models (LLMs)
- Best AI reasoning models
- Best AI coding models
- Best small and on-device AI models
- Best OCR and document AI models
- Best AI image models
- Best AI video models
- Best AI decision models
- Best text-to-speech models
- Best speech-to-text models
- Best AI music and audio models
- Best embedding models
- Best open-source LLMs
- Best American open-source LLMs
- Best Chinese open-source LLMs
- Best European open-source LLMs