Alternatives to EmbeddingGemma 2
Compare EmbeddingGemma 2 alternatives for your business or organization using the curated list below. SourceForge ranks the best alternatives to EmbeddingGemma 2 in 2026. Compare features, ratings, user reviews, pricing, and more from EmbeddingGemma 2 competitors and alternatives in order to make an informed decision for your business.
-
1
FLUX 3
Black Forest Labs
FLUX 3 is a multimodal foundation model that jointly learns from images, video, and audio within one unified architecture, building a representation of how objects hold together, how things move, and how events sound. Built on the Self-Flow approach, it aligns multimodal generation and understanding in the same backbone so each modality constrains the others, sound matches impact, motion follows physical properties, and future events follow from the past. FLUX 3 can mix modalities and jointly generate images, video, and native audio from text prompts or references such as images, video, and audio. Its video capabilities include text-to-video, image-to-video animation, video-to-video transformation, generative video-and-audio continuation, keyframe-controlled transitions, multilingual dialogue, animated typography, diverse styles and aspect ratios, and agentic chaining into longer multi-shot sequences. -
2
EmbeddingGemma
Google
EmbeddingGemma is a 308-million-parameter multilingual text embedding model, lightweight yet powerful, optimized to run entirely on everyday devices such as phones, laptops, and tablets, enabling fast, offline embedding generation that protects user privacy. Built on the Gemma 3 architecture, it supports over 100 languages, processes up to 2,000 input tokens, and leverages Matryoshka Representation Learning (MRL) to offer flexible embedding dimensions (768, 512, 256, or 128) for tailored speed, storage, and precision. Its GPU-and EdgeTPU-accelerated inference delivers embeddings in milliseconds, under 15 ms for 256 tokens on EdgeTPU, while quantization-aware training keeps memory usage under 200 MB without compromising quality. This makes it ideal for real-time, on-device tasks such as semantic search, retrieval-augmented generation (RAG), classification, clustering, and similarity detection, whether for personal file search, mobile chatbots, or custom domain use. -
3
Nomic Embed
Nomic
Nomic Embed is a suite of open source, high-performance embedding models designed for various applications, including multilingual text, multimodal content, and code. The ecosystem includes models like Nomic Embed Text v2, which utilizes a Mixture-of-Experts (MoE) architecture to support over 100 languages with efficient inference using 305M active parameters. Nomic Embed Text v1.5 offers variable embedding dimensions (64 to 768) through Matryoshka Representation Learning, enabling developers to balance performance and storage needs. For multimodal applications, Nomic Embed Vision v1.5 aligns with the text models to provide a unified latent space for text and image data, facilitating seamless multimodal search. Additionally, Nomic Embed Code delivers state-of-the-art performance on code embedding tasks across multiple programming languages.Starting Price: Free -
4
Cohere Embed
Cohere
Cohere's Embed is a leading multimodal embedding platform designed to transform text, images, or a combination of both into high-quality vector representations. These embeddings are optimized for semantic search, retrieval-augmented generation, classification, clustering, and agentic AI applications. The latest model, embed-v4.0, supports mixed-modality inputs, allowing users to combine text and images into a single embedding. It offers Matryoshka embeddings with configurable dimensions of 256, 512, 1024, or 1536, enabling flexibility in balancing performance and resource usage. With a context length of up to 128,000 tokens, embed-v4.0 is well-suited for processing large documents and complex data structures. It also supports compressed embedding types, including float, int8, uint8, binary, and ubinary, facilitating efficient storage and faster retrieval in vector databases. Multilingual support spans over 100 languages, making it a versatile tool for global applications.Starting Price: $0.47 per image -
5
voyage-code-3
MongoDB
Voyage AI introduces voyage-code-3, a next-generation embedding model optimized for code retrieval. It outperforms OpenAI-v3-large and CodeSage-large by an average of 13.80% and 16.81% on a suite of 32 code retrieval datasets, respectively. It supports embeddings of 2048, 1024, 512, and 256 dimensions and offers multiple embedding quantization options, including float (32-bit), int8 (8-bit signed integer), uint8 (8-bit unsigned integer), binary (bit-packed int8), and ubinary (bit-packed uint8). With a 32 K-token context length, it surpasses OpenAI's 8K and CodeSage Large's 1K context lengths. Voyage-code-3 employs Matryoshka learning to create embeddings with a nested family of various lengths within a single vector. This allows users to vectorize documents into a 2048-dimensional vector and later use shorter versions (e.g., 256, 512, or 1024 dimensions) without re-invoking the embedding model. -
6
txtai
NeuML
txtai is an all-in-one open source embeddings database designed for semantic search, large language model orchestration, and language model workflows. It unifies vector indexes (both sparse and dense), graph networks, and relational databases, providing a robust foundation for vector search and serving as a powerful knowledge source for LLM applications. With txtai, users can build autonomous agents, implement retrieval augmented generation processes, and develop multi-modal workflows. Key features include vector search with SQL support, object storage integration, topic modeling, graph analysis, and multimodal indexing capabilities. It supports the creation of embeddings for various data types, including text, documents, audio, images, and video. Additionally, txtai offers pipelines powered by language models that handle tasks such as LLM prompting, question-answering, labeling, transcription, translation, and summarization.Starting Price: Free -
7
Gemini Embedding
Google
Gemini Embedding’s first text model (gemini-embedding-001) is now generally available via the Gemini API and Gemini Enterprise Agent Platform, having held a top spot on the Massive Text Embedding Benchmark Multilingual leaderboard since its experimental launch in March, thanks to superior performance across retrieval, classification, and other embedding tasks compared to both legacy Google and external proprietary models. Exceptionally versatile, it supports over 100 languages with a 2,048‑token input limit and employs the Matryoshka Representation Learning (MRL) technique to let developers choose output dimensions of 3072, 153,6, or 768 for optimal quality, performance, and storage efficiency.Starting Price: $0.15 per 1M input tokens -
8
voyage-3-large
MongoDB
Voyage AI has unveiled voyage-3-large, a cutting-edge general-purpose and multilingual embedding model that leads across eight evaluated domains, including law, finance, and code, outperforming OpenAI-v3-large and Cohere-v3-English by averages of 9.74% and 20.71%, respectively. Enabled by Matryoshka learning and quantization-aware training, it supports embeddings of 2048, 1024, 512, and 256 dimensions, along with multiple quantization options such as 32-bit floating point, signed and unsigned 8-bit integer, and binary precision, significantly reducing vector database costs with minimal impact on retrieval quality. Notably, voyage-3-large offers a 32K-token context length, surpassing OpenAI's 8K and Cohere's 512 tokens. Evaluations across 100 datasets in diverse domains demonstrate its superior performance, with flexible precision and dimensionality options enabling substantial storage savings without compromising quality. -
9
Gemma 3n
Google DeepMind
Gemma 3n is our state-of-the-art open multimodal model, engineered for on-device performance and efficiency. Made for responsive, low-footprint local inference, Gemma 3n empowers a new wave of intelligent, on-the-go applications. It analyzes and responds to combined images and text, with video and audio coming soon. Build intelligent, interactive features that put user privacy first and work reliably offline. Mobile-first architecture, with a significantly reduced memory footprint. Co-designed by Google's mobile hardware teams and industry leaders. 4B active memory footprint with the ability to create submodels for quality-latency tradeoffs. Gemma 3n is our first open model built on this groundbreaking, shared architecture, allowing developers to begin experimenting with this technology today in an early preview. -
10
Marengo
TwelveLabs
Marengo is a multimodal video foundation model that transforms video, audio, image, and text inputs into unified embeddings, enabling powerful “any-to-any” search, retrieval, classification, and analysis across vast video and multimedia libraries. It integrates visual frames (with spatial and temporal dynamics), audio (speech, ambient sound, music), and textual content (subtitles, overlays, metadata) to create a rich, multidimensional representation of each media item. With this embedding architecture, Marengo supports robust tasks such as search (text-to-video, image-to-video, video-to-audio, etc.), semantic content discovery, anomaly detection, hybrid search, clustering, and similarity-based recommendation. The latest versions introduce multi-vector embeddings, separating representations for appearance, motion, and audio/text features, which significantly improve precision and context awareness, especially for complex or long-form content.Starting Price: $0.042 per minute -
11
Arctic Embed 2.0
Snowflake
Snowflake's Arctic Embed 2.0 introduces multilingual capabilities to its text embedding models, enhancing global-scale retrieval without compromising English performance or scalability. Building upon the robust foundation of previous releases, Arctic Embed 2.0 supports multiple languages, enabling developers to create stream-processing pipelines that incorporate neural networks and complex tasks like tracking, video encoding/decoding, and rendering, facilitating real-time analytics on various data types. The model leverages Matryoshka Representation Learning (MRL) for efficient embedding storage, allowing for significant compression with minimal quality degradation. This advancement ensures that enterprises can handle demanding workloads such as training large-scale models, fine-tuning, real-time inference, and high-performance computing tasks across diverse languages and regions.Starting Price: $2 per credit -
12
MedGemma
Google DeepMind
MedGemma is a collection of Gemma 3 variants that are trained for performance on medical text and image comprehension. Developers can use MedGemma to accelerate building healthcare-based AI applications. MedGemma currently comes in two variants: a 4B multimodal version and a 27B text-only version. MedGemma 4B utilizes a SigLIP image encoder that has been specifically pre-trained on a variety of de-identified medical data, including chest X-rays, dermatology images, ophthalmology images, and histopathology slides. Its LLM component is trained on a diverse set of medical data, including radiology images, histopathology patches, ophthalmology images, and dermatology images. MedGemma 4B is available in both pre-trained (suffix: -pt) and instruction-tuned (suffix -it) versions. The instruction-tuned version is a better starting point for most applications. -
13
Gemini Embedding 2
Google
Gemini Embedding models, including the newer Gemini Embedding 2, are part of Google’s Gemini AI ecosystem and are designed to convert text, phrases, sentences, and code into numerical vector representations that capture their semantic meaning. Unlike generative models that produce new content, the embedding model transforms input data into dense vectors that represent meaning in a mathematical format, allowing computers to compare and analyze information based on conceptual similarity rather than exact wording. These embeddings enable applications such as semantic search, recommendation systems, document retrieval, clustering, classification, and retrieval-augmented generation pipelines. The model can process input in more than 100 languages and supports up to 2048 tokens per request, allowing it to embed longer pieces of text or code while maintaining strong contextual understanding.Starting Price: Free -
14
voyage-4-large
Voyage AI
The Voyage 4 model family from Voyage AI is a new generation of text embedding models designed to produce high-quality semantic vectors with an industry-first shared embedding space that lets different models in the series generate compatible embeddings so developers can mix and match models for document and query embedding to optimize accuracy, latency, and cost trade-offs. It includes voyage-4-large (a flagship model using a mixture-of-experts architecture delivering state-of-the-art retrieval accuracy at about 40% lower serving cost than comparable dense models), voyage-4 (balancing quality and efficiency), voyage-4-lite (high-quality embeddings with fewer parameters and lower compute cost), and the open-weight voyage-4-nano (ideal for local development and prototyping with an Apache 2.0 license). All four models in the series operate in a single shared embedding space, so embeddings generated by different variants are interchangeable, enabling asymmetric retrieval strategies. -
15
Universal Sentence Encoder
Tensorflow
The Universal Sentence Encoder (USE) encodes text into high-dimensional vectors that can be utilized for tasks such as text classification, semantic similarity, and clustering. It offers two model variants: one based on the Transformer architecture and another on Deep Averaging Network (DAN), allowing a balance between accuracy and computational efficiency. The Transformer-based model captures context-sensitive embeddings by processing the entire input sequence simultaneously, while the DAN-based model computes embeddings by averaging word embeddings, followed by a feedforward neural network. These embeddings facilitate efficient semantic similarity calculations and enhance performance on downstream tasks with minimal supervised training data. The USE is accessible via TensorFlow Hub, enabling seamless integration into various applications. -
16
fastText
fastText
fastText is an open source, free, and lightweight library developed by Facebook's AI Research (FAIR) lab for efficient learning of word representations and text classification. It supports both unsupervised learning of word vectors and supervised learning for text classification tasks. A key feature of fastText is its ability to capture subword information by representing words as bags of character n-grams, which enhances the handling of morphologically rich languages and out-of-vocabulary words. The library is optimized for performance and capable of training on large datasets quickly, and the resulting models can be reduced in size for deployment on mobile devices. Pre-trained word vectors are available for 157 languages, trained on Common Crawl and Wikipedia data, and can be downloaded for immediate use. fastText also offers aligned word vectors for 44 languages, facilitating cross-lingual natural language processing tasks.Starting Price: Free -
17
GloVe
Stanford NLP
GloVe (Global Vectors for Word Representation) is an unsupervised learning algorithm developed by the Stanford NLP Group to obtain vector representations for words. It constructs word embeddings by analyzing global word-word co-occurrence statistics from a given corpus, resulting in vector spaces where the geometric relationships reflect semantic similarities and differences among words. A notable feature of GloVe is its ability to capture linear substructures within the word vector space, enabling vector arithmetic to express relationships. The model is trained on the non-zero entries of a global word-word co-occurrence matrix, which records how frequently pairs of words appear together in a corpus. This approach efficiently leverages statistical information by focusing on significant co-occurrences, leading to meaningful word representations. Pre-trained word vectors are available for various corpora, including Wikipedia 2014.Starting Price: Free -
18
NVIDIA NeMo Retriever
NVIDIA
NVIDIA NeMo Retriever is a collection of microservices for building multimodal extraction, reranking, and embedding pipelines with high accuracy and maximum data privacy. It delivers quick, context-aware responses for AI applications like advanced retrieval-augmented generation (RAG) and agentic AI workflows. As part of the NVIDIA NeMo platform and built with NVIDIA NIM, NeMo Retriever allows developers to flexibly leverage these microservices to connect AI applications to large enterprise datasets wherever they reside and fine-tune them to align with specific use cases. NeMo Retriever provides components for building data extraction and information retrieval pipelines. The pipeline extracts structured and unstructured data (e.g., text, charts, tables), converts it to text, and filters out duplicates. A NeMo Retriever embedding NIM converts the chunks into embeddings and stores them in a vector database, accelerated by NVIDIA cuVS, for enhanced performance and speed of indexing. -
19
Mistral Small 3.1
Mistral
Mistral Small 3.1 is a state-of-the-art, multimodal, and multilingual AI model released under the Apache 2.0 license. Building upon Mistral Small 3, this enhanced version offers improved text performance, and advanced multimodal understanding, and supports an expanded context window of up to 128,000 tokens. It outperforms comparable models like Gemma 3 and GPT-4o Mini, delivering inference speeds of 150 tokens per second. Designed for versatility, Mistral Small 3.1 excels in tasks such as instruction following, conversational assistance, image understanding, and function calling, making it suitable for both enterprise and consumer-grade AI applications. Its lightweight architecture allows it to run efficiently on a single RTX 4090 or a Mac with 32GB RAM, facilitating on-device deployments. It is available for download on Hugging Face, accessible via Mistral AI's developer playground, and integrated into platforms likeGemini Enterprise Agent Platform, with availability on NVIDIA NIM.Starting Price: Free -
20
E5 Text Embeddings
Microsoft
E5 Text Embeddings, developed by Microsoft, are advanced models designed to convert textual data into meaningful vector representations, enhancing tasks like semantic search and information retrieval. These models are trained using weakly-supervised contrastive learning on a vast dataset of over one billion text pairs, enabling them to capture intricate semantic relationships across multiple languages. The E5 family includes models of varying sizes—small, base, and large—offering a balance between computational efficiency and embedding quality. Additionally, multilingual versions of these models have been fine-tuned to support diverse languages, ensuring broad applicability in global contexts. Comprehensive evaluations demonstrate that E5 models achieve performance on par with state-of-the-art, English-only models of similar sizes.Starting Price: Free -
21
word2vec
Google
Word2Vec is a neural network-based technique for learning word embeddings, developed by researchers at Google. It transforms words into continuous vector representations in a multi-dimensional space, capturing semantic relationships based on context. Word2Vec uses two main architectures: Skip-gram, which predicts surrounding words given a target word, and Continuous Bag-of-Words (CBOW), which predicts a target word based on surrounding words. By training on large text corpora, Word2Vec generates word embeddings where similar words are positioned closely, enabling tasks like semantic similarity, analogy solving, and text clustering. The model was influential in advancing NLP by introducing efficient training techniques such as hierarchical softmax and negative sampling. Though newer embedding models like BERT and Transformer-based methods have surpassed it in complexity and performance, Word2Vec remains a foundational method in natural language processing and machine learning research.Starting Price: Free -
22
Mixedbread
Mixedbread
Mixedbread is a fully-managed AI search engine that allows users to build production-ready AI search and Retrieval-Augmented Generation (RAG) applications. It offers a complete AI search stack, including vector stores, embedding and reranking models, and document parsing. Users can transform raw data into intelligent search experiences that power AI agents, chatbots, and knowledge systems without the complexity. It integrates with tools like Google Drive, SharePoint, Notion, and Slack. Its vector stores enable users to build production search engines in minutes, supporting over 100 languages. Mixedbread's embedding and reranking models have achieved over 50 million downloads and outperform OpenAI in semantic search and RAG tasks while remaining open-source and cost-effective. The document parser extracts text, tables, and layouts from PDFs, images, and complex documents, providing clean, AI-ready content without manual preprocessing. -
23
Voyage AI
MongoDB
Voyage AI provides best-in-class embedding models and rerankers designed to supercharge search and retrieval for unstructured data. Its technology powers high-quality Retrieval-Augmented Generation (RAG) by improving how relevant context is retrieved before responses are generated. Voyage AI offers general-purpose, domain-specific, and company-specific models to support a wide range of use cases. The models are optimized for accuracy, low latency, and reduced costs through shorter vector dimensions. With long-context support of up to 32K tokens, Voyage AI enables deeper understanding of complex documents. The platform is modular and integrates easily with any vector database or large language model. Voyage AI is trusted by industry leaders to deliver reliable, factual AI outputs at scale. -
24
Neum AI
Neum AI
No one wants their AI to respond with out-of-date information to a customer. Neum AI helps companies have accurate and up-to-date context in their AI applications. Use built-in connectors for data sources like Amazon S3 and Azure Blob Storage, vector stores like Pinecone and Weaviate to set up your data pipelines in minutes. Supercharge your data pipeline by transforming and embedding your data with built-in connectors for embedding models like OpenAI and Replicate, and serverless functions like Azure Functions and AWS Lambda. Leverage role-based access controls to make sure only the right people can access specific vectors. Bring your own embedding models, vector stores and sources. Ask us about how you can even run Neum AI in your own cloud. -
25
BGE
BGE
BGE (BAAI General Embedding) is a comprehensive retrieval toolkit designed for search and Retrieval-Augmented Generation (RAG) applications. It offers inference, evaluation, and fine-tuning capabilities for embedding models and rerankers, facilitating the development of advanced information retrieval systems. The toolkit includes components such as embedders and rerankers, which can be integrated into RAG pipelines to enhance search relevance and accuracy. BGE supports various retrieval methods, including dense retrieval, multi-vector retrieval, and sparse retrieval, providing flexibility to handle different data types and retrieval scenarios. The models are available through platforms like Hugging Face, and the toolkit provides tutorials and APIs to assist users in implementing and customizing their retrieval systems. By leveraging BGE, developers can build robust and efficient search solutions tailored to their specific needs.Starting Price: Free -
26
BilberryDB
BilberryDB
BilberryDB is an enterprise-grade vector-database platform designed for building AI applications that handle multimodal data, including images, video, audio, 3D models, tabular data, and text, across one unified system. It supports lightning-fast similarity search and retrieval via embeddings, allows few-shot or no-code workflows to create powerful search/classification capabilities without large labelled datasets, and offers a developer SDK (such as TypeScript) as well as a visual builder for non-technical users. The platform emphasises sub-second query performance at scale, seamless ingestion of diverse data types, and rapid deployment of vector-search-enabled apps (“Deploy as an App”) so organisations can build AI-driven search, recommendation, classification, or content-discovery systems without building infrastructure from scratch.Starting Price: Free -
27
Codestral Embed
Mistral AI
Codestral Embed is Mistral AI's first embedding model, specialized for code, optimized for high-performance code retrieval and semantic understanding. It significantly outperforms leading code embedders in the market today, such as Voyage Code 3, Cohere Embed v4.0, and OpenAI’s large embedding model. Codestral Embed can output embeddings with different dimensions and precisions; for instance, with a dimension of 256 and int8 precision, it still performs better than any model from competitors. The dimensions of the embeddings are ordered by relevance, allowing users to choose the first n dimensions for a smooth trade-off between quality and cost. It excels in retrieval use cases on real-world code data, particularly in benchmarks like SWE-Bench, which is based on real-world GitHub issues and corresponding fixes, and Text2Code (GitHub), relevant for providing context for code completion or editing. -
28
LexVec
Alexandre Salle
LexVec is a word embedding model that achieves state-of-the-art results in multiple natural language processing tasks by factorizing the Positive Pointwise Mutual Information (PPMI) matrix using stochastic gradient descent. This approach assigns heavier penalties for errors on frequent co-occurrences while accounting for negative co-occurrences. Pre-trained vectors are available, including a common crawl dataset with 58 billion tokens and 2 million words in 300 dimensions, and an English Wikipedia 2015 + NewsCrawl dataset with 7 billion tokens and 368,999 words in 300 dimensions. Evaluations demonstrate that LexVec matches or outperforms other models like word2vec in terms of word similarity and analogy tasks. The implementation is open source under the MIT License and is available on GitHub.Starting Price: Free -
29
DiffusionGemma
Google
DiffusionGemma is an experimental open model that explores text diffusion, an exceptionally fast approach to text generation. Released under an Apache 2.0 license, this 26B Mixture of Experts (MoE) model moves beyond the sequential token-by-token processing of typical autoregressive Large Language Models (LLMs). Instead, it generates entire blocks of text simultaneously, delivering up to 4x faster text generation on GPUs. Built on the intelligence-per-parameter of the Gemma 4 family and Gemini Diffusion research, DiffusionGemma integrates a novel diffusion head designed to maximize generation speed. It is designed for researchers and developers exploring speed-critical, interactive local workflows such as in-line editing, rapid iteration, and non-linear text structures. By shifting the decode bottleneck from memory bandwidth to compute, it can generate more than 1,000 tokens per second on a single NVIDIA H100 and more than 700 tokens per second on an NVIDIA GeForce RTX 5090.Starting Price: Free -
30
Jina AI
Jina AI
Empower businesses and developers to create cutting-edge neural search, generative AI, and multimodal services using state-of-the-art LMOps, MLOps and cloud-native technologies. Multimodal data is everywhere: from simple tweets to photos on Instagram, short videos on TikTok, audio snippets, Zoom meeting records, PDFs with figures, 3D meshes in games. It is rich and powerful, but that power often hides behind different modalities and incompatible data formats. To enable high-level AI applications, one needs to solve search and create first. Neural Search uses AI to find what you need. A description of a sunrise can match a picture, or a photo of a rose can match a song. Generative AI/Creative AI uses AI to make what you need. It can create an image from a description, or write poems from a picture. -
31
PaliGemma 2
Google
PaliGemma 2, the next evolution in tunable vision-language models, builds upon the performant Gemma 2 models, adding the power of vision and making it easier than ever to fine-tune for exceptional performance. With PaliGemma 2, these models can see, understand, and interact with visual input, opening up a world of new possibilities. It offers scalable performance with multiple model sizes (3B, 10B, 28B parameters) and resolutions (224px, 448px, 896px). PaliGemma 2 generates detailed, contextually relevant captions for images, going beyond simple object identification to describe actions, emotions, and the overall narrative of the scene. Our research demonstrates leading performance in chemical formula recognition, music score recognition, spatial reasoning, and chest X-ray report generation, as detailed in the technical report. Upgrading to PaliGemma 2 is a breeze for existing PaliGemma users. -
32
Cloudflare Vectorize
Cloudflare
Begin building for free in minutes. Vectorize enables fast & cost-effective vector storage to power your search & AI Retrieval Augmented Generation (RAG) applications. Avoid tool sprawl & reduce total cost of ownership, Vectorize seamlessly integrates with Cloudflare’s AI developer platform and AI gateway for centralized development, monitoring & control of AI applications on a global scale. Vectorize is a globally distributed vector database that enables you to build full-stack, AI-powered applications with Cloudflare Workers AI. Vectorize makes querying embeddings, representations of values or objects like text, images, and audio that are designed to be consumed by machine learning models and semantic search algorithms, faster, easier, and more affordable. Search, similarity, recommendation, classification & anomaly detection based on your own data. Improved results & faster search. String, number & boolean types are supported. -
33
LFM2.5
Liquid AI
Liquid AI’s LFM2.5 is the next generation of on-device AI foundation models designed to deliver high-performance, efficient AI inference on edge devices such as phones, laptops, vehicles, IoT systems, and embedded hardware without relying on cloud compute. It extends the previous LFM2 architecture by significantly increasing the pretraining scale and reinforcement learning stages, yielding a family of hybrid models around 1.2 billion parameters that balance instruction following, reasoning, and multimodal capabilities for real-world agentic use cases. The LFM2.5 family includes Base (for fine-tuning and customization), Instruct (general-purpose instruction-tuned), Japanese-optimized, Vision-Language, and Audio-Language variants, all optimized for fast, on-device inference under tight memory constraints and available as open-weight models deployable via frameworks like llama.cpp, MLX, vLLM, and ONNX.Starting Price: Free -
34
Gemma 2
Google
A family of state-of-the-art, light-open models created from the same research and technology that were used to create Gemini models. These models incorporate comprehensive security measures and help ensure responsible and reliable AI solutions through selected data sets and rigorous adjustments. Gemma models achieve exceptional comparative results in their 2B, 7B, 9B, and 27B sizes, even outperforming some larger open models. With Keras 3.0, enjoy seamless compatibility with JAX, TensorFlow, and PyTorch, allowing you to effortlessly choose and change frameworks based on task. Redesigned to deliver outstanding performance and unmatched efficiency, Gemma 2 is optimized for incredibly fast inference on various hardware. The Gemma family of models offers different models that are optimized for specific use cases and adapt to your needs. Gemma models are large text-to-text lightweight language models with a decoder, trained in a huge set of text data, code, and mathematical content. -
35
Context Data
Context Data
Context Data is an enterprise data infrastructure built to accelerate the development of data pipelines for Generative AI applications. The platform automates the process of setting up internal data processing and transformation flows using an easy-to-use connectivity framework where developers and enterprises can quickly connect to all of their internal data sources, embedding models and vector database targets without having to set up expensive infrastructure or engineers. The platform also allows developers to schedule recurring data flows for refreshed and up-to-date data.Starting Price: $99 per month -
36
ZeroEntropy
ZeroEntropy
ZeroEntropy is a search and retrieval platform built to deliver faster, more accurate, human-level search experiences. It provides cutting-edge rerankers, embeddings, and hybrid retrieval models that go beyond traditional lexical and vector search. ZeroEntropy focuses on understanding context, nuance, and domain-specific meaning rather than just keywords. Its models consistently outperform leading alternatives on industry benchmarks. Developers can integrate ZeroEntropy quickly using a simple, production-ready API. The platform is optimized for low latency, high accuracy, and cost efficiency. ZeroEntropy enables teams to ship search systems that actually return the right answers. -
37
Qwen3-VL
Alibaba
Qwen3-VL is the newest vision-language model in the Qwen family (by Alibaba Cloud), designed to fuse powerful text understanding/generation with advanced visual and video comprehension into one unified multimodal model. It accepts inputs in mixed modalities, text, images, and video, and handles long, interleaved contexts natively (up to 256 K tokens, with extensibility beyond). Qwen3-VL delivers major advances in spatial reasoning, visual perception, and multimodal reasoning; the model architecture incorporates several innovations such as Interleaved-MRoPE (for robust spatio-temporal positional encoding), DeepStack (to leverage multi-level features from its Vision Transformer backbone for refined image-text alignment), and text–timestamp alignment (for precise reasoning over video content and temporal events). These upgrades enable Qwen3-VL to interpret complex scenes, follow dynamic video sequences, read and reason about visual layouts.Starting Price: Free -
38
LocalAI
LocalAI
LocalAI is a free, open source, local-first AI platform designed as a drop-in replacement for the OpenAI API, allowing developers to run large language models and other AI systems entirely on their own hardware without relying on cloud services. It provides a complete AI stack for local inferencing, enabling text generation, image creation with diffusion models, audio transcription and speech synthesis, embeddings for semantic search, and multimodal capabilities such as vision analysis. It is compatible with OpenAI API specifications, allowing existing applications to integrate seamlessly by simply switching endpoints, while supporting a wide range of open source model families that can run on CPU or GPU, including consumer-grade devices. LocalAI emphasizes privacy and control by ensuring all processing happens locally, keeping data on-device and eliminating external dependencies.Starting Price: Free -
39
Mu
Microsoft
Mu is a 330-million-parameter encoder–decoder language model designed to power the agent in Windows settings by mapping natural-language queries to Settings function calls, running fully on-device via NPUs at over 100 tokens per second while maintaining high accuracy. Drawing on Phi Silica optimizations, Mu’s encoder–decoder architecture reuses a fixed-length latent representation to cut computation and memory overhead, yielding 47 percent lower first-token latency and 4.7× higher decoding speed on Qualcomm Hexagon NPUs compared to similar decoder-only models. Hardware-aware tuning, including a 2/3–1/3 encoder–decoder parameter split, weight sharing between input and output embeddings, Dual LayerNorm, rotary positional embeddings, and grouped-query attention, enables fast inference at over 200 tokens per second on devices like Surface Laptop 7 and sub-500 ms response times for settings queries. -
40
MiMo-V2.5
Xiaomi Technology
Xiaomi MiMo-V2.5 is an advanced open-source AI model designed to combine strong agentic capabilities with native multimodal understanding. It can process and reason across text, images, and audio within a single unified system. The model uses a sparse Mixture-of-Experts architecture with hundreds of billions of parameters for efficient performance. It supports an extended context window of up to one million tokens, enabling long and complex workflows. MiMo-V2.5 is built to handle tasks such as coding, reasoning, and multimodal analysis with high accuracy. It incorporates dedicated visual and audio encoders to enhance perception and cross-modal reasoning. The model demonstrates strong benchmark performance across coding, reasoning, and multimodal tasks. By combining multimodality, efficiency, and agentic intelligence, MiMo-V2.5 advances the capabilities of open-source AI systems. -
41
GLM-4.6V
Z.ai
GLM-4.6V is a state-of-the-art open source multimodal vision-language model from the Z.ai (GLM-V) family designed for reasoning, perception, and action. It ships in two variants: a full-scale version (106B parameters) for cloud or high-performance clusters, and a lightweight “Flash” variant (9B) optimized for local deployment or low-latency use. GLM-4.6V supports a native context window of up to 128K tokens during training, enabling it to process very long documents or multimodal inputs. Crucially, it integrates native Function Calling, meaning the model can take images, screenshots, documents, or other visual media as input directly (without manual text conversion), reason about them, and trigger tool calls, bridging “visual perception” with “executable action.” This enables a wide spectrum of capabilities; interleaved image-and-text content generation (for example, combining document understanding with text summarization or generation of image-annotated responses).Starting Price: Free -
42
TranslateGemma
Google
TranslateGemma is a new suite of open machine translation models from Google built on the Gemma 3 foundation that lets people and systems communicate across 55 languages with high-quality AI translation while maintaining efficiency and broad deployment flexibility. Available in 4 B, 12 B, and 27 B parameter sizes, TranslateGemma distills advanced multilingual capabilities into compact models that can run on mobile devices, consumer laptops, local machines, or cloud hardware without sacrificing accuracy or performance; technical evaluations show the 12 B version can outperform larger baseline models with lower compute demands. The models were developed through a specialized two-stage fine-tuning process combining high-quality human and synthetic translation data with reinforcement learning to optimize translation quality across diverse language families.Starting Price: Free -
43
Gensim
Radim Řehůřek
Gensim is a free, open source Python library designed for unsupervised topic modeling and natural language processing, focusing on large-scale semantic modeling. It enables the training of models like Word2Vec, FastText, Latent Semantic Analysis (LSA), and Latent Dirichlet Allocation (LDA), facilitating the representation of documents as semantic vectors and the discovery of semantically related documents. Gensim is optimized for performance with highly efficient implementations in Python and Cython, allowing it to process arbitrarily large corpora using data streaming and incremental algorithms without loading the entire dataset into RAM. It is platform-independent, running on Linux, Windows, and macOS, and is licensed under the GNU LGPL, promoting both personal and commercial use. The library is widely adopted, with thousands of companies utilizing it daily, over 2,600 academic citations, and more than 1 million downloads per week.Starting Price: Free -
44
DataGemma
Google
DataGemma represents a pioneering effort by Google to enhance the accuracy and reliability of large language models (LLMs) when dealing with statistical and numerical data. Launched as a set of open models, DataGemma leverages Google's Data Commons, a vast repository of public statistical data—to ground its responses in real-world facts. This initiative employs two innovative approaches: Retrieval Interleaved Generation (RIG) and Retrieval Augmented Generation (RAG). The RIG method integrates real-time data checks during the generation process to ensure factual accuracy, while RAG retrieves relevant information before generating responses, thereby reducing the likelihood of AI hallucinations. By doing so, DataGemma aims to provide users with more trustworthy and factually grounded answers, marking a significant step towards mitigating the issue of misinformation in AI-generated content. -
45
Marqo
Marqo
Marqo is more than a vector database, it's an end-to-end vector search engine. Vector generation, storage, and retrieval are handled out of the box through a single API. No need to bring your own embeddings. Accelerate your development cycle with Marqo. Index documents and begin searching in just a few lines of code. Create multimodal indexes and search combinations of images and text with ease. Choose from a range of open source models or bring your own. Build interesting and complex queries with ease. With Marqo you can compose queries with multiple weighted components. With Marqo, input pre-processing, machine learning inference, and storage are all included out of the box. Run Marqo in a Docker image on your laptop or scale it up to dozens of GPU inference nodes in the cloud. Marqo can be scaled to provide low-latency searches against multi-terabyte indexes. Marqo helps you configure deep-learning models like CLIP to pull semantic meaning from images.Starting Price: $86.58 per month -
46
Gemma
Google
Gemma is a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Developed by Google DeepMind and other teams across Google, Gemma is inspired by Gemini, and the name reflects the Latin gemma, meaning “precious stone.” Accompanying our model weights, we’re also releasing tools to support developer innovation, foster collaboration, and guide the responsible use of Gemma models. Gemma models share technical and infrastructure components with Gemini, our largest and most capable AI model widely available today. This enables Gemma 2B and 7B to achieve best-in-class performance for their sizes compared to other open models. And Gemma models are capable of running directly on a developer laptop or desktop computer. Notably, Gemma surpasses significantly larger models on key benchmarks while adhering to our rigorous standards for safe and responsible outputs. -
47
Muse Glimmer
Meta
Muse Glimmer is a 30-billion-parameter open-weights model from Meta Superintelligence Labs, optimized for always-on local agent workflows. Small enough to run on a Mac or PC with a single consumer GPU, it is designed for local agents, function calling, coding, and LLM-as-a-judge evaluation without depending on cloud infrastructure or network access. The model combines long-horizon execution, precise tool calling, multimodal understanding, long-context memory, and instruction following. It can complete end-to-end agentic tasks, sustain multi-step reasoning across extended workflows, recover from failed or unexpected tool calls, and accept interleaved text and images through a dedicated perception encoder for interpreting screenshots, charts, and documents. Muse Glimmer works with OpenClaw and other agentic orchestration patterns, supports controllable reasoning effort, and is trained on data from more than 100 languages.Starting Price: Free -
48
Exa
Exa.ai
The Exa API retrieves the best content on the web using embeddings-based search. Exa understands meaning, giving results search engines can’t. Exa uses a novel link prediction transformer to predict links which match the meaning of a prompt. For queries that need semantic understanding, search with our SOTA web embeddings model over our custom index. For all other queries, we offer keyword-based search. Stop learning how to web scrape or parse HTML. Get the clean, full text of any page in our index, or intelligent embeddings-ranked highlights related to a query. Select any date range, include or exclude any domain, select a custom data vertical, or get up to 10 million results..Starting Price: $100 per month -
49
Google AI Edge Gallery
Google
Google AI Edge Gallery is an experimental, open source Android app that demonstrates on-device machine learning and generative AI use cases, letting users download and run models locally (so they work offline once installed). It offers several features including AI Chat (multi-turn conversation), Ask Image (upload or use images to ask questions, identify objects, get descriptions), Audio Scribe (transcribe or translate recorded/uploaded audio), Prompt Lab (for single-turn tasks such as summarization, rewriting, code generation), and performance insights (metrics like latency, decode speed, etc.). Users can switch between different compatible models (including Gemma 3n and models from Hugging Face), bring their own LiteRT models, and explore model cards and source code for transparency. The app aims to protect privacy by doing all processing on the device, no internet connection needed for core operations after models are loaded, reducing latency, and enhancing data security.Starting Price: Free -
50
Llama 3.2
Meta
The open-source AI model you can fine-tune, distill and deploy anywhere is now available in more versions. Choose from 1B, 3B, 11B or 90B, or continue building with Llama 3.1. Llama 3.2 is a collection of large language models (LLMs) pretrained and fine-tuned in 1B and 3B sizes that are multilingual text only, and 11B and 90B sizes that take both text and image inputs and output text. Develop highly performative and efficient applications from our latest release. Use our 1B or 3B models for on device applications such as summarizing a discussion from your phone or calling on-device tools like calendar. Use our 11B or 90B models for image use cases such as transforming an existing image into something new or getting more information from an image of your surroundings.Starting Price: Free
