Alternatives to LFM3
Compare LFM3 alternatives for your business or organization using the curated list below. SourceForge ranks the best alternatives to LFM3 in 2026. Compare features, ratings, user reviews, pricing, and more from LFM3 competitors and alternatives in order to make an informed decision for your business.
-
1
LTX
Lightricks
LTX is an open foundation model for video, audio, and world simulation. You get full control over your AI: run LTX locally on your own hardware, fine tune it on your own IP, and generate video and audio as one unified output instead of stitching together separate tools. The latest model, LTX-2.5, is a 22B-parameter dual-stream diffusion transformer that generates native 4K video at up to 50fps, with synchronized audio and video produced in a single pass. Weights, code, and research are fully open, and independent benchmarking from Artificial Analysis ranks LTX among the top 3 AI video models globally. Access LTX three ways: download the open weights and run it yourself, license the model for on-premise deployment with enterprise support, or build on LTX Studio, the production suite for creative teams and studios. Teams at ElevenLabs, Asteria Film Co., Magnopus, and NVIDIA already build on LTX. LTX is production infrastructure for AI teams generating motion and physical environments inside their own pipelines, not a consumer app for one-off clips. -
2
Muse Spark Open
Meta
Muse Spark Open is an open-weight, natively multimodal language model for developers and organizations building AI applications, coding systems, and agentic workflows. The released model has 816 billion total parameters, 42 billion active parameters, and a 1-million-token context window. Teams can download the model weights, run inference through supported engines such as vLLM, SGLang, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, or Unsloth, and customize the model using PyTorch. It supports long-context reasoning, software engineering, coding, tool use, and multimodal inputs. Deployment options include cloud inference through listed API providers or self-hosting on compatible GPU systems; Meta documents a desktop setup using two Apple M5 Ultra Mac Studio computers. The model is released under the Muse Model License with limited commercial terms.Starting Price: $0 -
3
Nemotron 3.5 Bolt
NVIDIA
NVIDIA Nemotron 3.5 Bolt is a family of open artificial intelligence models designed for local AI agents, autonomous task execution, and conversational applications. The family includes Nemotron 3.5 Bolt 74B-A9B, a 74-billion-parameter mixture-of-experts model with approximately 9 billion active parameters, and Nemotron 3.5 Bolt E4B, a compact model optimized for conversational agents. The 74B-A9B variant uses native 2-bit quantization to operate within approximately 24GB of memory on compatible NVIDIA hardware. Its agentic capabilities support web research, software development, code debugging, and multistep task automation without requiring cloud-based inference. The E4B variant targets local voice assistants, interactive characters, and embedded applications with a 4GB memory footprint. Nemotron 3.5 Bolt supports deployment on NVIDIA RTX systems, DGX Spark, and other compatible hardware, giving developers greater control over inference, application data, and local AI workflows.Starting Price: Free -
4
Nemotron 3.5 Super VL
NVIDIA
NVIDIA Nemotron 3.5 Super VL is an open, customizable multimodal AI model designed for complex reasoning, autonomous agents, and enterprise workflows. The model uses a 120-billion-parameter mixture-of-experts architecture that activates approximately 12 billion parameters per token to balance computational efficiency with model capabilities. It combines text, image, document, and video understanding with instruction following, coding, and external tool use. Its multimodal capabilities support document analysis, chart interpretation, optical character recognition, video understanding, and automated information extraction. Developers can customize the model through supervised fine-tuning, low-rank adaptation, and reinforcement learning using NVIDIA NeMo. Nemotron 3.5 Super VL supports deployment in local, cloud, and data center environments for applications such as software engineering, financial document processing, cybersecurity analysis, and enterprise automation.Starting Price: $0 -
5
MiMo-V2.6-Pro
Xiaomi Technology
MiMo-V2.6-Pro is Xiaomi MiMo’s most capable open-source omnimodal AI model, built for coding, general agent workflows, visual tasks, research, and multimodal creation. The model combines strong software engineering capabilities with computer use, 3D spatial reasoning, visual perception, and tool use for complex multi-step work. MiMo-V2.6-Pro can build interactive 3D environments, generate Blender models, create frontend interfaces and presentations, and coordinate agents to refine outputs through visual feedback. It also supports research workflows such as literature review, computational experimentation, materials discovery, and formal mathematical proof development. Xiaomi trained the model with large-scale reinforcement learning across coding, general agents, visual tasks, and cybersecurity environments and has open-sourced the technical report, training environments, and RL code.Starting Price: Free -
6
MiMo-V2.6-Flash
Xiaomi Technology
MiMo-V2.6-Flash is an open-source, natively omnimodal AI model from Xiaomi MiMo designed to balance intelligence, efficiency, and cost. The model supports coding, general agent workflows, visual reasoning, computer use, automation, and multimodal creative tasks. Its capabilities extend beyond software engineering into frontend design, presentation creation, 3D modeling, interactive world generation, video production, and embodied simulation. Xiaomi trained MiMo-V2.6-Flash with large-scale reinforcement learning across coding, general agent, visual, and cybersecurity tasks, completing roughly 750,000 training trajectories. The model is positioned as the more cost-efficient member of the MiMo-V2.6 family while retaining strong performance across software engineering, tool use, automation, and visual coding benchmarks. MiMo-V2.6-Flash is available through MiMo Desktop, AI Studio, MiMo Code, the Xiaomi MiMo API Platform, OpenRouter, and the open-source MiMo-V2.6 release.Starting Price: Free -
7
GLM-5.3
Z.ai
GLM-5.3 is Z.ai’s frontier coding model designed for complex software engineering, long-horizon agent tasks, and advanced post-training research. The model uses the same base model as GLM-5.2, with improvements coming from scaled post-training across more environments, more diverse tasks, and larger compute investment. GLM-5.3 delivers stronger coding performance, better task ownership, improved benchmark results, and greater efficiency across realistic development workflows. It is built to handle complex coding tasks, production-style engineering work, research environments, automation tasks, and agentic workflows that require multi-step execution. The model also shows emergent cyber capabilities in vulnerability discovery and exploitation-chain reasoning, with safety evaluation and hardening planned before open-weight release.Starting Price: Free -
8
Qwen3.8-Max
Alibaba
Qwen3.8-Max is Qwen’s most capable model to date, built as a Max-class AI model for coding, work, research, long-horizon tasks, and multimodal agents. It scales to 2.4 trillion parameters with 95 billion active parameters and is available through QwenCloud. The model is designed to complete complex, open-ended tasks end to end with greater reliability and minimal human involvement. Qwen3.8-Max supports autonomous coding workflows, agentic development, research reproduction, visual reasoning, document understanding, video analysis, and real-world productivity tasks. It can integrate with popular agent frameworks and coding assistants, including Claude Code, Codex, Qoder CLI, Qwen Code, and OpenClaw. Built for developers, researchers, enterprises, and AI agent builders, Qwen3.8-Max helps teams automate sophisticated work across code, documents, tools, interfaces, and multimodal content.Starting Price: $2 per 1M (input) -
9
Inkling
Thinking Machines Lab
Inkling is an open-weights multimodal AI model from Thinking Machines designed as a customizable foundation model for developers, researchers, and enterprises. The model is a Mixture-of-Experts transformer with 975 billion total parameters, 41 billion active parameters, and support for context windows up to 1 million tokens. Inkling was trained from scratch on text, images, audio, and video, giving it native capabilities across reasoning, coding, agentic tool use, vision, audio, factuality, and instruction following. It is built with controllable thinking effort so users can balance performance, latency, and token efficiency for different workloads. The model is available for fine-tuning on Tinker, with playground access, API availability through ecosystem partners, and full weights published on Hugging Face. Built for customization, Inkling gives teams an open-weights base model for building domain-specific AI systems, multimodal agents, coding workflows, research tools, and more.Starting Price: Free -
10
MiniMax M3
MiniMax
MiniMax M3 is an open-weight multimodal AI model designed for coding, agentic workflows, long-context reasoning, and complex automation tasks. The model combines frontier-level coding performance, native multimodal understanding, and a context window of up to 1 million tokens. MiniMax M3 uses MiniMax Sparse Attention to improve long-context efficiency while reducing compute requirements for large-scale inputs. It supports text, image, and video understanding, making it useful for workflows that combine code, documents, visual references, and tool-driven tasks. The model is built for repository-scale reasoning, software engineering, autonomous task execution, tool calling, and multi-step agent workflows. MiniMax M3 helps developers, AI teams, and enterprises build capable agents that can reason across large contexts and work with multimodal information.Starting Price: $0.30 per million input tokens -
11
GLM-5.3-Flash
Z.ai
GLM-5.3-Flash is Z.ai’s natively multimodal model in the GLM-5 series (previously previewed as Ox Alpha), designed to deliver strong coding, agentic, visual, and knowledge-work performance at relatively low inference cost. It uses 320 billion total parameters with 18 billion active parameters, along with a hybrid architecture that combines sparse and linear attention to reduce the cost of long-context processing. The model supports context lengths of up to one million tokens and was trained on a 30-trillion-token multimodal corpus. GLM-5.3-Flash can reason across text, images, documents, interfaces, dashboards, and other visual information while using that feedback to refine its own outputs. Z.ai reports substantial gains over GLM-5.2 on coding and agentic benchmarks, including DeepSWE and AutomationBench, while approaching higher-cost frontier models on several evaluations.Starting Price: $0.15 per 1M tokens (input) -
12
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 -
13
Gemma 4
Google
Gemma 4 is an AI model introduced by Google and built on the Gemini architecture to deliver improved performance and flexibility. The model is designed to run efficiently on a single GPU or TPU, making it more accessible to developers and researchers. Gemma 4 enhances capabilities in natural language understanding and text generation, supporting a wide range of AI-driven applications. Its architecture allows it to handle complex tasks while maintaining efficient resource usage. Developers can use the model to build applications that rely on advanced language processing and automation. The design emphasizes scalability so that it can support both smaller projects and larger AI systems. By combining efficiency with powerful language capabilities, Gemma 4 helps advance the development of modern AI solutions.Starting Price: Free -
14
Qwen3.5
Alibaba
Qwen3.5 is a next-generation open-weight multimodal large language model designed to power native vision-language agents. The flagship release, Qwen3.5-397B-A17B, combines a hybrid linear attention architecture with sparse mixture-of-experts, activating only 17 billion parameters per forward pass out of 397 billion total to maximize efficiency. It delivers strong benchmark performance across reasoning, coding, multilingual understanding, visual reasoning, and agent-based tasks. The model expands language support from 119 to 201 languages and dialects while introducing a 1M-token context window in its hosted version, Qwen3.5-Plus. Built for multimodal tasks, it processes text, images, and video with advanced spatial reasoning and tool integration. Qwen3.5 also incorporates scalable reinforcement learning environments to improve general agent capabilities. Designed for developers and enterprises, it enables efficient, tool-augmented, multimodal AI workflows.Starting Price: Free -
15
Qwen3.8-27B
Alibaba
Qwen3.8-27B is a compact open-weights model in Alibaba’s Qwen3.8 family, aimed at developers and researchers who want strong local AI performance without using the full Max-scale model. Reports from Alibaba’s Qwen3.8 launch state that Qwen3.8-27B was planned for open-weight release alongside Qwen3.8-Max, expanding access for builders working on AI applications. The model is positioned for coding, research, professional workflows, and local deployment scenarios where a 27B model can be more practical than frontier-scale systems. Qwen3.8’s broader launch emphasizes software development, document processing, data analysis, and professional “cowork” use cases. Qwen3.8-27B is especially relevant for teams that need a capable open model for experimentation, coding agents, assistant workflows, and self-hosted inference. Built for practical deployment, Qwen3.8-27B gives developers a smaller Qwen3.8 option for building AI tools, testing agents, and running advanced language model workflows. -
16
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 -
17
DeepSeek-V4.1-Flash-1017
DeepSeek
DeepSeek-V4.1-Flash-1017 is an open-weight multimodal artificial intelligence model developed by DeepSeek for advanced reasoning, software engineering, visual understanding, and autonomous AI agent workflows. Released on October 17, 2026, it builds on DeepSeek-V4.1-Flash with improvements in coding performance, agentic task execution, and reasoning efficiency. The model supports text and image inputs, tool calling, configurable reasoning effort, and multistep problem-solving. Its one-million-token context window enables developers to process lengthy documents, extensive codebases, and complex sequences of instructions. DeepSeek-V4.1-Flash-1017 also supports end-to-end software development, including application generation, debugging, automated testing, and interactive project creation. The model is available through the DeepSeek API and downloadable weights under the MIT License, supporting both hosted and self-managed deployment.Starting Price: Free -
18
Mistral Large 3
Mistral AI
Mistral Large 3 is a next-generation, open multimodal AI model built with a powerful sparse Mixture-of-Experts architecture featuring 41B active parameters out of 675B total. Designed from scratch on NVIDIA H200 GPUs, it delivers frontier-level reasoning, multilingual performance, and advanced image understanding while remaining fully open-weight under the Apache 2.0 license. The model achieves top-tier results on modern instruction benchmarks, positioning it among the strongest permissively licensed foundation models available today. With native support across vLLM, TensorRT-LLM, and major cloud providers, Mistral Large 3 offers exceptional accessibility and performance efficiency. Its design enables enterprise-grade customization, letting teams fine-tune or adapt the model for domain-specific workflows and proprietary applications. Mistral Large 3 represents a major advancement in open AI, offering frontier intelligence without sacrificing transparency or control.Starting Price: Free -
19
Qwen3.8-Flash-Next
Alibaba
Qwen3.8-Flash-Next is an open-weight multimodal Mixture-of-Experts model and an early preview of the architecture planned for Qwen4. It systematically upgrades attention, residual connections, embeddings, and optimization to improve capability, computational efficiency, model capacity, and training stability. Its hybrid architecture combines Gated DeltaNet, which efficiently compresses historical information, with Qwen Sparse Attention, which selects important context at the micro-block level to reduce attention and indexing costs on long sequences. Gated Residual widens the residual stream into four branches and dynamically controls information flow across layers, while N-gram Embedding adds large-scale local-pattern memory with very little extra per-token computation and can be offloaded to host memory. The model uses a 125B-parameter main network plus 51B N-gram embedding parameters, while activating only 6B parameters per token.Starting Price: $2 per 1M (input) -
20
Laguna XS.2
Poolside
Laguna XS.2 is Poolside’s open-weight agentic coding model, built as the lightest and fastest model in the Laguna family. It is a 33B total-parameter Mixture of Experts model with 3B activated parameters, trained completely in-house on 30T tokens. As Poolside’s newest generation model open to the community, Laguna XS.2 is a second-generation architecture and the company’s first open-weight model, built on the lessons learned from training Laguna M.1 across synthetic data and reinforcement learning. The model is designed for agentic coding workflows, where it can code, act, iterate quickly, and perform best inside Poolside’s coding agent. Laguna XS.2 is positioned as a strong model for rapid agentic iteration, especially for developers and teams that need a compact, efficient coding model rather than a heavier frontier system. It is released under an Apache 2.0 license, allowing the community to evaluate, fine-tune, quantize, serve, and build on the weights.Starting Price: Free -
21
EXAONE Deep
LG
EXAONE Deep is a series of reasoning-enhanced language models developed by LG AI Research, featuring parameter sizes of 2.4 billion, 7.8 billion, and 32 billion. These models demonstrate superior capabilities in various reasoning tasks, including math and coding benchmarks. Notably, EXAONE Deep 2.4B outperforms other models of comparable size, EXAONE Deep 7.8B surpasses both open-weight models of similar scale and the proprietary reasoning model OpenAI o1-mini, and EXAONE Deep 32B shows competitive performance against leading open-weight models. The repository provides comprehensive documentation covering performance evaluations, quickstart guides for using EXAONE Deep models with Transformers, explanations of quantized EXAONE Deep weights in AWQ and GGUF formats, and instructions for running EXAONE Deep models locally using frameworks like llama.cpp and Ollama.Starting Price: Free -
22
Qwen3.6
Alibaba
Qwen3.6 is a large language model developed by Alibaba as part of its Qwen AI model family, designed for real-world applications and advanced reasoning tasks. It focuses on improving stability, usability, and performance compared to earlier versions. The model supports multimodal capabilities, allowing it to process and reason across text, images, and other data types. Qwen3.6 is particularly strong in coding and developer workflows, offering improved accuracy for complex programming tasks. It uses a mixture-of-experts architecture, enabling efficient performance while maintaining large-scale model capabilities. The model is designed to be deployable in production environments, including enterprise and cloud-based systems. It can be integrated into applications or run locally using open-weight variants. Overall, Qwen3.6 delivers a powerful, efficient, and versatile AI solution for modern use cases.Starting Price: Free -
23
Pixtral Large
Mistral AI
Pixtral Large is a 124-billion-parameter open-weight multimodal model developed by Mistral AI, building upon their Mistral Large 2 architecture. It integrates a 123-billion-parameter multimodal decoder with a 1-billion-parameter vision encoder, enabling advanced understanding of documents, charts, and natural images while maintaining leading text comprehension capabilities. With a context window of 128,000 tokens, Pixtral Large can process at least 30 high-resolution images simultaneously. The model has demonstrated state-of-the-art performance on benchmarks such as MathVista, DocVQA, and VQAv2, surpassing models like GPT-4o and Gemini-1.5 Pro. Pixtral Large is available under the Mistral Research License for research and educational use, and under the Mistral Commercial License for commercial applications.Starting Price: Free -
24
Phonon
Gradium
Phonon is an open-weight text-to-speech (TTS) model and inference runtime for developers adding speech generation to applications that need local, offline audio. It converts text into spoken audio and can stream speech as it is generated. The runtime provides Rust and Python interfaces, a Swift package for iOS and macOS, JavaScript for browsers, and Android libraries; it can also run as a self-hosted server with an OpenAI-compatible API. Depending on the model variant, it supports English, French, German, Spanish, and Portuguese. Phonon is intended for developers integrating speech into mobile, desktop, web, embedded, and server applications, including offline or privacy-sensitive use cases. The 31-million-parameter English model weights use the Apache 2.0 license; 40-million-parameter language-specific and multilingual weights use CC BY-NC 4.0. The runtime code is separately available under Apache 2.0 or MIT. -
25
Nemotron 3.5 Transcribe
NVIDIA
NVIDIA Nemotron 3.5 Transcribe is an open-weight speech and audio model for developers building transcription and audio-processing systems. It converts recordings into text and supports verbatim or edited transcription, contextual vocabulary, speaker attribution, word-level timestamps, speech translation, language identification, and audio question answering. Its language matrix covers 40 languages, with task coverage varying by language. The model can process recordings up to one hour and is designed for offline inference using compatible NVIDIA GPU hardware and supported runtimes. Teams can integrate it into speech interfaces, meeting and media processing, customer-service analytics, translation tools, and other audio-understanding applications. Model weights and a fine-tuning toolkit are released under the OpenMDW-1.1 license; hardware and serving requirements depend on deployment. -
26
Tülu 3
Ai2
Tülu 3 is an advanced instruction-following language model developed by the Allen Institute for AI (Ai2), designed to enhance capabilities in areas such as knowledge, reasoning, mathematics, coding, and safety. Built upon the Llama 3 Base, Tülu 3 employs a comprehensive four-stage post-training process: meticulous prompt curation and synthesis, supervised fine-tuning on a diverse set of prompts and completions, preference tuning using both off- and on-policy data, and a novel reinforcement learning approach to bolster specific skills with verifiable rewards. This open-source model distinguishes itself by providing full transparency, including access to training data, code, and evaluation tools, thereby closing the performance gap between open and proprietary fine-tuning methods. Evaluations indicate that Tülu 3 outperforms other open-weight models of similar size, such as Llama 3.1-Instruct and Qwen2.5-Instruct, across various benchmarks.Starting Price: Free -
27
Qwen2
Alibaba
Qwen2 is the large language model series developed by Qwen team, Alibaba Cloud. Qwen2 is a series of large language models developed by the Qwen team at Alibaba Cloud. It includes both base language models and instruction-tuned models, ranging from 0.5 billion to 72 billion parameters, and features both dense models and a Mixture-of-Experts model. The Qwen2 series is designed to surpass most previous open-weight models, including its predecessor Qwen1.5, and to compete with proprietary models across a broad spectrum of benchmarks in language understanding, generation, multilingual capabilities, coding, mathematics, and reasoning.Starting Price: Free -
28
Fastino
Fastino
Fastino is an applied AI platform focused on specialized, open-weight language models and the Fastino Fine-Tuning Agent. The agent lets users describe a task in plain language, then automatically selects the architecture, generates training data, trains and evaluates the model, and returns a task-specific model ready to deploy. Fine-tuning projects can be created and revisited from one interface, with models trained on the user’s terms and deployable in their own environment. The resulting models are designed for production-grade performance, with typical response times under 50 ms, while model weights remain private and owned by the user. Models can move from a task description to a trained result in hours, giving teams a faster path to specialized deployment. Fastino also provides open-source and open-weight models for specialized AI workloads. -
29
Codestral
Mistral AI
We introduce Codestral, our first-ever code model. Codestral is an open-weight generative AI model explicitly designed for code generation tasks. It helps developers write and interact with code through a shared instruction and completion API endpoint. As it masters code and English, it can be used to design advanced AI applications for software developers. Codestral is trained on a diverse dataset of 80+ programming languages, including the most popular ones, such as Python, Java, C, C++, JavaScript, and Bash. It also performs well on more specific ones like Swift and Fortran. This broad language base ensures Codestral can assist developers in various coding environments and projects.Starting Price: Free -
30
GLM-5.1
Z.ai
GLM-5.1 is the latest iteration of Z.ai’s GLM series, designed as a frontier-level, agent-oriented AI model optimized for coding, reasoning, and long-horizon workflows. It builds on the GLM-5 architecture, which uses a Mixture-of-Experts (MoE) design to deliver high performance while keeping inference costs efficient, and is part of a broader push toward open-weight, developer-accessible models. A core focus of GLM-5.1 is enabling agentic behavior, meaning it can plan, execute, and iterate across multi-step tasks rather than simply responding to single prompts. It is specifically designed to handle complex workflows such as debugging code, navigating repositories, and executing chained operations with sustained context. Compared to earlier models, GLM-5.1 improves reliability in long interactions, maintaining coherence across extended sessions and reducing breakdowns in multi-step reasoning.Starting Price: Free -
31
Mixtral 8x7B
Mistral AI
Mixtral 8x7B is a high-quality sparse mixture of experts model (SMoE) with open weights. Licensed under Apache 2.0. Mixtral outperforms Llama 2 70B on most benchmarks with 6x faster inference. It is the strongest open-weight model with a permissive license and the best model overall regarding cost/performance trade-offs. In particular, it matches or outperforms GPT-3.5 on most standard benchmarks.Starting Price: Free -
32
FLUX 3 Action
Black Forest Labs
FLUX 3 Action is an open-weight 7B world-action model designed for action prediction in robotics and other latency-sensitive visual environments. Derived from the multimodal FLUX 3 backbone, it was pretrained on large-scale image, video, and audio data with a strong emphasis on video, then adapted through joint video-action training and fine-tuning for specific embodiments and action spaces. Given an instruction, camera observations, and robot joint positions, the model predicts motor commands together with their expected visual outcomes, allowing a robot to execute actions, observe the environment again, and continuously replan. Unlike approaches that separate visual prediction from control, FLUX 3 Action jointly models future video and actions, transferring world understanding learned from broad video pretraining into robot control. Its single-step 7B checkpoint reaches a 38.3% success rate on RoboLab-120. -
33
Shieldstral
Mistral AI
Shieldstral is a 3B open-weights, policy-adaptive multimodal safety classifier designed to evaluate text, images, and text-plus-image content using policies defined at inference time. Instead of relying on a fixed taxonomy of harm categories, it frames moderation as a binary question-answering task: users provide an instruction describing the evaluation context and strictness, a yes-or-no safety question, and the content to judge. The model reads the “yes” and “no” logits and converts them into a continuous, calibrated safety score, allowing applications to threshold or rank results by confidence rather than depend on a single discrete label. This formulation unifies prompt classification, response moderation, refusal detection, toxicity detection, and multimodal safety in one interface, while letting teams adapt policies without retraining the model. Shieldstral can evaluate prompts, responses, prompt-response pairs, images, and images with accompanying text. -
34
Olmo 2
Ai2
Olmo 2 is a family of fully open language models developed by the Allen Institute for AI (AI2), designed to provide researchers and developers with transparent access to training data, open-source code, reproducible training recipes, and comprehensive evaluations. These models are trained on up to 5 trillion tokens and are competitive with leading open-weight models like Llama 3.1 on English academic benchmarks. Olmo 2 emphasizes training stability, implementing techniques to prevent loss spikes during long training runs, and utilizes staged training interventions during late pretraining to address capability deficiencies. The models incorporate state-of-the-art post-training methodologies from AI2's Tülu 3, resulting in the creation of Olmo 2-Instruct models. An actionable evaluation framework, the Open Language Modeling Evaluation System (OLMES), was established to guide improvements through development stages, consisting of 20 evaluation benchmarks assessing core capabilities. -
35
Beam
Reflection
Beam is Reflection’s first open-weight model, a sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active parameters, built for coding, reasoning, and agentic workloads. Its capabilities come from large-scale pretraining and reinforcement learning, including training on 23.8 trillion diverse, curated, high-quality tokens from the web, public sources, and proprietary licensed datasets. Beam was trained with a particular focus on coding and agentic performance and is designed to deliver competitive open-weight capabilities with efficient inference compute. It supports complex software engineering, terminal, STEM, web search, tool-use, and general knowledge tasks, with reinforcement learning designed to improve multi-step reasoning, tool use, and adaptation to environment feedback. Users can control the tradeoff between performance and token usage through a reasoning effort parameter. -
36
Kimi K2.5
Moonshot AI
Kimi K2.5 is a next-generation multimodal AI model designed for advanced reasoning, coding, and visual understanding tasks. It features a native multimodal architecture that supports both text and visual inputs, enabling image and video comprehension alongside natural language processing. Kimi K2.5 delivers open-source state-of-the-art performance in agent workflows, software development, and general intelligence tasks. The model offers ultra-long context support with a 256K token window, making it suitable for large documents and complex conversations. It includes long-thinking capabilities that allow multi-step reasoning and tool invocation for solving challenging problems. Kimi K2.5 is fully compatible with the OpenAI API format, allowing developers to switch seamlessly with minimal changes. With strong performance, flexibility, and developer-focused tooling, Kimi K2.5 is built for production-grade AI applications.Starting Price: Free -
37
Antares
Cisco
Antares is a family of open-weight security small language models purpose-built to localize known vulnerabilities inside large codebases. Antares-350M and Antares-1B are compact enough to run locally or on premises, helping teams keep proprietary source code inside their environment while reducing inference cost and runtime. Starting from a vulnerability description, advisory, or CWE category, the model follows an iterative investigation process similar to a human analyst, it searches for relevant code patterns, reads candidate files, incorporates new evidence, changes direction when a path is unproductive, and narrows the search to the files most likely to contain the weakness. Antares returns a ranked list of potentially vulnerable source files together with the terminal exploration trace that produced the result, making findings easier to review and prioritize. -
38
Ideogram 4.0
Ideogram
Ideogram 4.0 is an open image model at the forefront of design, built for open weights, multilingual text, precise layout control, editable elements, and realistic 2K images. It is a state-of-the-art open-weight image model for developers and enterprises that want to build, fine-tune, and run visual intelligence on their own hardware. Ideogram 4.0 was trained with a describe-to-structure-to-recreate loop, first reading scenes, backgrounds, text, and objects as structured data, then learning to rebuild images from that representation. This approach is designed to help the model understand composition before recreating it, giving teams more control over layout, objects, typography, and visual structure. It is built for real design work, especially brand, advertising, fashion, marketing, food, apparel, social, photography, and illustration use cases. Ideogram has led on text rendering since launch, and 4.0 adds bounding-box layout control so headlines stay readable.Starting Price: Free -
39
Mistral Large 4
Mistral AI
Mistral Large 4 is a general-purpose multimodal language model available through Mistral’s hosted inference API. It combines instruction following, reasoning and agentic capabilities, and accepts image and text inputs. Developers can use it for chat completion, structured outputs, function calling, document question answering, tool-assisted workflows and other application features. Its documented use cases include conversational assistants, coding and software engineering agents, cybersecurity analysis, finance and legal workflows, scientific research, data extraction, summarization, and analysis of documents, charts and images. Mistral Large 4 is designed for developers, researchers, businesses and public-sector teams building AI-powered applications across industries such as engineering, manufacturing, finance, logistics and science.Starting Price: $1.36 per 1M input tokens -
40
GLM-4.5V-Flash
Z.ai
GLM-4.5V-Flash is an open source vision-language model, designed to bring strong multimodal capabilities into a lightweight, deployable package. It supports image, video, document, and GUI inputs, enabling tasks such as scene understanding, chart and document parsing, screen reading, and multi-image analysis. Compared to larger models in the series, GLM-4.5V-Flash offers a compact footprint while retaining core VLM capabilities like visual reasoning, video understanding, GUI task handling, and complex document parsing. It can serve in “GUI agent” workflows, meaning it can interpret screenshots or desktop captures, recognize icons or UI elements, and assist with automated desktop or web-based tasks. Although it forgoes some of the largest-model performance gains, GLM-4.5V-Flash remains versatile for real-world multimodal tasks where efficiency, lower resource usage, and broad modality support are prioritized.Starting Price: Free -
41
Nemotron 3 Nano Omni
NVIDIA
NVIDIA Nemotron 3 Nano Omni is an open, omni-modal foundation model designed to unify perception and reasoning across text, images, audio, video, and documents within a single efficient architecture. It eliminates the need for separate models for each modality, reducing inference latency, orchestration complexity, and cost while maintaining consistent cross-modal context. It is purpose-built for agentic AI systems, acting as a perception and context sub-agent that gives larger AI agents the ability to “see, hear, and read” in real time across screens, recordings, and structured or unstructured data. It supports advanced multimodal reasoning tasks such as document understanding, speech recognition, long audio-video analysis, and computer-use workflows, enabling agents to interpret dynamic interfaces and complex environments. Built with a hybrid architecture optimized for long context and throughput, it can process large inputs like multi-page documents.Starting Price: Free -
42
Ling 3.0 Tiny
Ant Group
Ling 3.0 Tiny is an open-weights reasoning model with 7.9B total parameters, 1.3B active parameters, and a 262K-token context window. Built with a mixture-of-experts architecture, it extends the open-weights Pareto frontier for intelligence versus active parameters and is small enough to run locally in many settings. The model scores 25 on the Artificial Analysis Intelligence Index, comparable to gpt-oss-120b (high, 24) while using 15x fewer total parameters and 4x fewer active parameters. This parameter efficiency comes with relatively high token usage, with 213M output tokens required to run the Intelligence Index. Ling 3.0 Tiny also shows substantial improvements in hallucination behavior over Ling-mini-2.0, improving its AA-Omniscience score by 59 points while maintaining similar accuracy. Rather than guessing when uncertain, it attempted only 37% of questions in the evaluation, resulting in a 30% hallucination rate compared with 96% for the previous generation. -
43
Llama 2
Meta
The next generation of our open source large language model. This release includes model weights and starting code for pretrained and fine-tuned Llama language models — ranging from 7B to 70B parameters. Llama 2 pretrained models are trained on 2 trillion tokens, and have double the context length than Llama 1. Its fine-tuned models have been trained on over 1 million human annotations. Llama 2 outperforms other open source language models on many external benchmarks, including reasoning, coding, proficiency, and knowledge tests. Llama 2 was pretrained on publicly available online data sources. The fine-tuned model, Llama-2-chat, leverages publicly available instruction datasets and over 1 million human annotations. We have a broad range of supporters around the world who believe in our open approach to today’s AI — companies that have given early feedback and are excited to build with Llama 2.Starting Price: Free -
44
Phi-4-reasoning
Microsoft
Phi-4-reasoning is a 14-billion parameter transformer-based language model optimized for complex reasoning tasks, including math, coding, algorithmic problem solving, and planning. Trained via supervised fine-tuning of Phi-4 on carefully curated "teachable" prompts and reasoning demonstrations generated using o3-mini, it generates detailed reasoning chains that effectively leverage inference-time compute. Phi-4-reasoning incorporates outcome-based reinforcement learning to produce longer reasoning traces. It outperforms significantly larger open-weight models such as DeepSeek-R1-Distill-Llama-70B and approaches the performance levels of the full DeepSeek-R1 model across a wide range of reasoning tasks. Phi-4-reasoning is designed for environments with constrained computing or latency. Fine-tuned with synthetic data generated by DeepSeek-R1, it provides high-quality, step-by-step problem solving. -
45
Qwen3.6-27B
Alibaba
Qwen3.6-27B is a dense, open source multimodal language model in the Qwen3.6 series, designed to deliver flagship-level performance in coding, reasoning, and agent-based workflows while maintaining a relatively efficient parameter size of 27 billion. It is positioned as a high-performance general model that “punches above its weight,” achieving results competitive with or superior to significantly larger models on key benchmarks, particularly in agentic coding tasks. It supports both thinking and non-thinking modes, allowing it to dynamically balance deep reasoning with fast responses depending on the task, and integrates capabilities across text and multimodal inputs such as images and video. Built as part of the Qwen3.6 family, the model emphasizes real-world usability, stability, and developer productivity, incorporating improvements driven by community feedback and practical deployment needs.Starting Price: Free -
46
Smaug Flash
Abacus.AI
Smaug Flash is a family of three open-weight models fine-tuned by Abacus.AI for production agentic workloads, with each model positioned at a different point on the capability–efficiency curve. The line is trained using human-curated real-world agentic traces combined with synthetic data grounded in difficult examples, producing gains in agentic coding, real-world tool use, automation, long-context reasoning, and instruction following. Smaug Flash, based on DeepSeek V4 Flash 0731, is the workhorse model for enterprise self-improving agents where speed, efficiency, and reliable agent performance need to coexist. It is specifically tuned to reduce the spins and confusion that can appear during long-context tool use while retaining the base model’s speed advantages. Smaug Mini, based on Qwen3.8 27B, targets multimodal use cases and smaller reasoning tasks in a more compact package, with stronger real-world agentic ability for one-off workflows. -
47
FLUX.2
Black Forest Labs
FLUX.2 is built for real production workflows, delivering high-quality visuals while maintaining character, product, and style consistency across multiple reference images. It handles structured prompts, brand-safe layouts, complex text rendering, and detailed logos with precision. The model supports multi-reference inputs, editing at up to 4 megapixels, and generates both photorealistic scenes and highly stylized compositions. With a focus on reliability, FLUX.2 processes real-world creative tasks—such as infographics, product shots, and UI mockups—with exceptional stability. It represents Black Forest Labs’ open-core approach, pairing frontier-level capability with open-weight models that invite experimentation. Across its variants, FLUX.2 provides flexible options for studios, developers, and researchers who need scalable, customizable visual intelligence. -
48
Devstral Small 2
Mistral AI
Devstral Small 2 is the compact, 24 billion-parameter variant of the new coding-focused model family from Mistral AI, released under the permissive Apache 2.0 license to enable both local deployment and API use. Alongside its larger sibling (Devstral 2), this model brings “agentic coding” capabilities to environments with modest compute: it supports a large 256K-token context window, enabling it to understand and make changes across entire codebases. On the standard code-generation benchmark (SWE-Bench Verified), Devstral Small 2 scores around 68.0%, placing it among open-weight models many times its size. Because of its reduced size and efficient design, Devstral Small 2 can run on a single GPU or even CPU-only setups, making it practical for developers, small teams, or hobbyists without access to data-center hardware. Despite its compact footprint, Devstral Small 2 retains key capabilities of larger models; it can reason across multiple files and track dependencies.Starting Price: Free -
49
MiniMax M2
MiniMax
MiniMax M2 is an open source foundation model built specifically for agentic applications and coding workflows, striking a new balance of performance, speed, and cost. It excels in end-to-end development scenarios, handling programming, tool-calling, and complex, long-chain workflows with capabilities such as Python integration, while delivering inference speeds of around 100 tokens per second and offering API pricing at just ~8% of the cost of comparable proprietary models. The model supports “Lightning Mode” for high-speed, lightweight agent tasks, and “Pro Mode” for in-depth full-stack development, report generation, and web-based tool orchestration; its weights are fully open source and available for local deployment with vLLM or SGLang. MiniMax M2 positions itself as a production-ready model that enables agents to complete independent tasks, such as data analysis, programming, tool orchestration, and large-scale multi-step logic at real organizational scale.Starting Price: $0.30 per million input tokens -
50
Qwen3.6-35B-A3B
Alibaba
Qwen3.5-35B-A3B is part of the Qwen3.5 “Medium” model series, designed as a highly efficient, multimodal foundation model that balances strong reasoning ability with practical deployment requirements. It uses a Mixture-of-Experts (MoE) architecture with 35 billion total parameters but activates only about 3 billion per token, allowing it to deliver performance comparable to much larger models while significantly reducing computational cost. The model integrates a hybrid attention mechanism that combines linear attention with standard attention layers, enabling efficient long-context processing and improved scalability for complex tasks. As a native vision-language model, it can process both text and visual inputs, supporting use cases such as multimodal reasoning, coding, and agent-based workflows. It is designed to function as a general-purpose “AI agent,” capable of planning, tool use, and structured problem solving rather than just conversational responses.Starting Price: Free
