For visual grounding, pointing, counting, and video understanding, evaluate Molmo 2; for a single model spanning text, images, audio, and video, consider Qwen2.5-Omni; for vision analysis, localization, and structured visual outputs, consider Qwen2.5-VL. None is established as a drop-in replacement for Liquid AI’s d1 decision interface: d1 is designed to return an answer in one forward pass without generating tokens, while these alternatives are broader multimodal models. Available publisher evaluations do not provide a common head-to-head test that identifies a winner for multimodal decisions.
What makes d1 different from a general multimodal model?
Liquid AI announced d1-3B and the experimental d1-omni-600M on October 7, 2026. Liquid describes d1 as a decision-model family: rather than generating a sequence of tokens, a d1 model returns an answer in one forward pass. d1-3B accepts text and images; d1-omni-600M accepts text with images or text with audio.
As an Amazon Associate I earn from qualifying purchases.
That output contract matters if your application expects a fast classification or decision rather than a conversational response. A vision-language model or omni model may handle the same input, but its generated answer may need prompting, a constrained output schema, or a separate decision layer to fit your application. “Multimodal” describes the inputs and capabilities; it does not by itself mean “decision model.”
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Which open-weight alternative fits your task?
| Model family | Evaluate it for | What differs from d1 |
|---|---|---|
| Molmo 2 4B, 8B, O-7B |
Image and video understanding, visual grounding, pointing, counting, tracking, dense captioning, and video question answering. | Ai2 positions it as a multimodal model family, not a single-pass decision interface. Its strengths are relevant when locating or describing visual evidence is part of the task. |
| Qwen2.5-Omni 3B, 7B |
Workflows combining text, images, audio, and video, including streaming text and natural-speech responses. | It is designed for perception and generation, including speech responses; that is a different output workflow from d1’s non-token decision answer. |
| Qwen2.5-VL 3B, 7B, 72B |
Vision-language tasks involving charts, layouts, video, object localization, and structured visual outputs. | It is vision-focused rather than a direct substitute benchmarked against d1’s decision interface. The cited 3B model card labels its license “qwen-research.” |
Choose Molmo 2 when visual grounding is central
Molmo 2 is the clearest candidate to investigate when a decision depends on where something is in an image or video, or on counting, tracking, and describing visual content. Ai2 calls the 4B model a compact workhorse and identifies 8B as its strongest overall video-understanding performer. Ai2 describes Molmo 2-O (7B) as “a fully open, end-to-end stack for research.” Those are Ai2’s characterizations of its own models; they do not establish that Molmo replaces d1’s decision behavior.
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
Choose Qwen2.5-Omni when modalities and generated responses matter
Qwen2.5-Omni is the broadest modality option in this shortlist: its project materials describe text, image, audio, and video input, with streaming text and speech output. Consider it when the application needs an integrated perception-and-response model, not only a compact decision answer.
Choose Qwen2.5-VL for vision analysis and structured output
Qwen2.5-VL is the vision-oriented choice when the workflow involves documents, charts, layouts, video, or locating objects and returning structured visual information. Its model family spans 3B, 7B, and 72B variants; the cited task-specific figures below are from the 3B model card, not a claim that all sizes perform identically.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
How to interpret the published benchmark figures
The numbers below come from model publishers’ evaluations. They measure different tasks and benchmark sets, so they are not a shared leaderboard and cannot determine which model makes the best multimodal decisions.
- d1-3B: Liquid AI reports 48.57 on Decision Index v0.2.1’s public split and says d1-3B led every model under 10B on that index, performing on par with Decider 35B-A3B. Liquid also reports that it does not publish results for the private vision split used in Decision Index v0.3, and says dedicated audio decision benchmarks remain an open problem.
- Qwen2.5-Omni: the Qwen team’s 2025 project evaluation reports an OmniBench average of 56.13% for 7B and 52.19% for 3B. These are OmniBench results, not comparable to d1’s Decision Index score.
- Qwen2.5-VL-3B: its model card reports 93.9 on DocVQA test, 77.1 on InfoVQA test, and 62.3 on MathVista test-mini. It reports 67.6/61.5 on VideoMME; those paired figures are presented in the model card’s evaluation table and should be read in that benchmark’s listed context, not as an overall vision score.
- d1 text benchmark bundle: Liquid reports means of 82.9 for d1-3B and 78.4 for d1-omni-600M across seven listed text benchmarks: SQuAD 2.0, Civil Comments, MASSIVE intent, PubMedQA, BoolQ, XNLI, and PAWS-X. This is a text-benchmark aggregate, not a multimodal result.
Liquid’s October 2026 release summary also reports one-question latency of 8 ms on an NVIDIA GeForce RTX 4090, 16 ms on Jetson AGX Thor, 26 ms on Jetson AGX Orin, and 50 ms on Jetson Orin Nano. These are Liquid-reported measurements, not independent tests. Its detailed latency table includes different one-question/image and packed-state scenarios, so the summary figures should not be treated as universal latency guarantees or compared with another model without matching the workload and measurement procedure.
Rank #3
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
How to select and test a candidate
Before choosing on the basis of a model card or benchmark, define the exact decision the application must make and test candidate models on the same held-out examples.
- Specify the input. Record whether each example contains text, an image, video, audio, or a combination, and keep the input format consistent across candidates.
- Specify the output contract. Decide whether the system needs a label or structured decision, an explanation, a grounded location, or generated speech. Check whether the model returns that form directly or needs prompting and downstream processing.
- Measure task quality on common examples. Use the same held-out data, task definition, and scoring rules for each model. Do not compare a score from one family’s benchmark with another family’s unrelated benchmark.
- Check grounding requirements. If the result must point to an object, track it, or cite a region or moment in video, verify that the model’s output supports that requirement rather than assuming a text answer is grounded.
- Measure deployment behavior. Test the intended model size, inference stack, hardware, and latency procedure. Parameter count alone does not establish the memory, throughput, or response time your deployment will achieve.
- Review the actual license and openness claims. Confirm that the specific weights and components may be used for your purpose. “Open-weight” does not, on its own, establish that training data, recipes, or every component are open.
Licensing and openness need model-specific checks
Liquid calls d1-3B and d1-omni-600M open-weight; that description alone does not establish that the training data, recipes, or all components are open. Ai2 describes Molmo 2-O (7B) as a fully open end-to-end stack for research and details open data for the family. The cited Qwen2.5-VL-3B model card labels its license “qwen-research.” Check current terms for the particular model and deployment before relying on an openness label or assuming commercial permission.
Model files, licenses, and inference support may change. The publisher materials summarized here do not establish the status of every hosted inference provider or every deployment setup.
Practical verdict
Start with Molmo 2 if visual grounding and video understanding define the problem, Qwen2.5-Omni if you need broad text, image, audio, and video capability with generated responses, or Qwen2.5-VL if the work is primarily vision analysis and structured visual output. Treat each as a candidate for your workflow, not a proven d1 replacement. The available publisher pages do not establish a same-task winner for multimodal decision-making, particularly for audio decisions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
