Best AI Models - Page 14
View:
-
1
Nex-N2-mini
Nex-AGI
Nex-N2-mini is an open source agentic model with Agentic Thinking, built for real-world productivity scenarios where fast instruction following, real-time tool execution, and cost-effective large-scale deployment matter. As part of the Nex-N2 family, it is designed to turn thinking into actions that are executable, verifiable, and iterable, rather than treating reasoning, tool use, and environment execution as separate capabilities. Nex-N2-mini uses the same unified Agentic Thinking framework as Nex-N2-Pro, connecting requirement understanding, task planning, code implementation, environmental feedback, evaluation, debugging, and continuous iteration into one closed loop. Its thinking paradigm stays consistent across search, coding, and agentic tool calling, following goal decomposition, state tracking, strategy adjustment, and self-verification, which is especially useful in mixed tasks where coding is interleaved with searches and tool calls.Starting Price: Free -
2
DeepSeek-OCR
DeepSeek
DeepSeek-OCR is an open source model for Contexts Optical Compression, built to explore the boundaries of visual-text compression and investigate the role of vision encoders from an LLM-centric viewpoint. It is designed to compress long contexts through optical 2D mapping, using DeepEncoder as the core engine and DeepSeek3B-MoE-A570M as the decoder. DeepEncoder maintains low activations under high-resolution input while achieving high compression ratios, keeping the number of vision tokens manageable for document understanding. The model supports OCR and document parsing workflows for images and PDFs, with inference through vLLM or Transformers. Users can run image OCR with streaming output, process PDFs with high concurrency, or run batch evaluation for benchmarks. DeepSeek-OCR can convert documents to Markdown, perform free OCR without layouts, parse figures, describe images in detail, and locate referenced text inside an image.Starting Price: Free -
3
Hy3
Tencent
Hy3 preview is Tencent Hy’s most intelligent model in the Hy series to date, built as a 295B-parameter Mixture-of-Experts model with 21B activated parameters, 3.8B MTP layer parameters, and support for up to a 256K token context window. As the first model trained on Tencent Hy’s rebuilt infrastructure, Hy3 preview is designed to improve real-world usability across complex reasoning, instruction following, context learning, coding, agent capabilities, and overall inference performance. It integrates both fast and slow thinking capabilities, allowing direct responses for simpler tasks and deeper reasoning for complex math, coding, and reasoning work. The model is built around well-rounded capabilities across long-context understanding, instruction following, tool use, and agent workflows, with evaluation focused not only on standard benchmarks but also on authentic business and development scenarios.Starting Price: Free -
4
Ornith-1.0
DeepReinforce
Ornith-1.0 is a self-improving family of models built specially for agentic coding tasks. It spans the full spectrum from compact 9B Dense models suitable for edge device deployment to 397B MoE frontier-scale models optimized for maximum performance, with variants including 9B Dense, 31B Dense, 35B MoE, and 397B MoE. Built on top of pretrained Gemma 4 and Qwen 3.5, Ornith-1.0 achieves state-of-the-art performance among open-source models of comparable size on coding benchmarks. Its key innovation is a self-improving training framework that learns to generate both solution rollouts and the task-specific scaffolds that guide those rollouts. Instead of relying on fixed, human-designed harnesses, Ornith-1.0 treats the scaffold as a learnable object that co-evolves with the policy, allowing the model to jointly optimize the orchestration and the final solution.Starting Price: Free -
5
TabFM
Google
TabFM is a zero-shot foundation model for tabular data, designed to simplify classification and regression workflows that traditionally require manual model training, hyperparameter tuning, and domain-specific feature engineering. Built specifically for tables, TabFM reframes tabular prediction as an in-context learning problem: instead of fitting a new supervised model to each dataset, it takes historical training examples and target testing rows together as one unified prompt, then interprets relationships between columns and rows at inference time. Because tables are two-dimensional and orderless, TabFM uses a hybrid architecture that combines alternating row and column attention, row compression, and a dedicated Transformer for in-context learning over compressed row embeddings. This design lets the model capture complex feature interactions and dependencies while keeping prediction computationally efficient for larger datasets.Starting Price: Free -
6
Leanstral 1.5
Mistral AI
Leanstral 1.5 is an Apache-2.0 licensed model for practical proof engineering in Lean 4, built to make formal verification more powerful and accessible. With 119B total parameters and only 6B active parameters, it delivers a major performance upgrade for theorem proving, agentic proof engineering, and real-world code verification. Leanstral 1.5 was trained through a three-stage process: mid-training, supervised fine-tuning, and reinforcement learning with CISPO. In the multiturn environment, the model receives a theorem statement, submits a proof, gets Lean compiler feedback, and refines its approach until the proof compiles or the budget is exhausted. In the code agent environment, Leanstral works like a developer in a raw filesystem: it edits files, runs bash commands, and uses the Lean language server to inspect goals, errors, and type information in real time.Starting Price: Free -
7
Reve 2.1
Reve
Reve 2.1 is a new foundation image model that makes a rapid leap in visual intelligence and world knowledge, just one month after Reve 2.0. It extends the same foundation of controllability, but sharpens it at every stage with intuitive prompt understanding, stronger foreign-text rendering, and more precise native 4K output. Reve 2.1 plans in finer detail, reasons more accurately about how elements relate, and renders results with greater precision at full 16-megapixel resolution. Built around the belief that images should be structured like code, with hierarchical layouts and controllable regions, the model brings layout planning directly into visual intelligence. It reasons about structure, hierarchy, and spatial relationships before rendering, making it stronger for dense scenes, intricate compositions, complicated visual instructions, and fine text. Reve 2.1 also supports precision editing, where every element is addressable and editable.Starting Price: $7.99 per month -
8
Gemini 3.5 Flash-Lite
Google
Gemini 3.5 Flash-Lite is Google’s fastest model in the Gemini 3.5 series, designed for low-latency tasks and high-throughput developer workflows such as agentic search, document processing, coding, and large-scale data analysis. It delivers 350 output tokens per second and significantly improves on previous Flash-Lite generations in both quality and agentic performance. Developers can configure its thinking level to match the workload: minimal or low thinking supports fast execution for high-volume tasks, while higher thinking levels enable more complex, multi-step subagent workflows. Built-in computer-use capabilities allow the model to interact reliably with digital environments across supported surfaces. Gemini 3.5 Flash-Lite also advances coding, long-context understanding, and real-world task execution, outperforming Gemini 3.1 Flash-Lite across key evaluations and even surpassing Gemini 3 Flash on several agentic and software-engineering benchmarks.Starting Price: $0.30 per 1M input tokens -
9
GPT-Realtime-2.1
OpenAI
GPT-Realtime-2.1 is OpenAI’s reasoning model with tool use for low-latency voice agents and complex speech-to-speech workflows. It updates GPT-Realtime-2 with improved alphanumeric recognition, silence and noise handling, and interruption behavior, helping applications understand spoken code, manage imperfect audio, and respond more naturally when users pause or talk over the agent. Developers can configure reasoning effort to balance deeper thinking against latency and output usage, while strong instruction following helps the model stay aligned with a defined role, tone, and workflow. It accepts and produces both audio and text, can take images as input, and supports function calling so an agent can retrieve information or perform actions during a conversation. The model has a 128,000-token context window, supports up to 32,000 output tokens, and includes reasoning-token support for extended interactions.Starting Price: $0.40 per cached input -
10
WorldClaw
Tencent
WorldClaw is a fully agentic, coarse-to-fine framework for generating large-scale, freely explorable, and editable 3D open worlds from open-ended text prompts. Rather than synthesizing an entire world in a single pass, it follows a global-to-regional construction strategy that maintains coherent spatial organization while adding rich local detail. Planning agents first translate a prompt into a structured scene specification covering regions, terrain, assets, materials, visual style, and spatial relationships. It builds a globally coherent terrain foundation using semantic layouts, reusable assets, generative or procedural materials, and region-aware height fields. For regions requiring greater detail, WorldClaw creates terrain-conditioned compositions, reconstructs editable textured meshes, and recovers their placement within the scene. Render-based agents further refine terrain geometry, objects, appearance, arrangements, and contacts with the environment.Starting Price: Free -
11
Happy Shrimp 1.0
Alibaba Cloud
Happy Shrimp 1.0 is an AI music generation model designed to turn vague ideas into fully produced songs from a single prompt. Users can start with an emotion, story concept, target genre, or creative direction, without needing technical music knowledge such as BPM, key signature, or instrumentation. The model supports text-to-music generation for complete vocal songs and instrumental compositions, producing melody, arrangement, lyrics, and vocals from scratch or using lyrics supplied by the user. Built with broad musical knowledge, it understands creative aesthetics across genres, regions, eras, and cultural contexts, translating descriptive prompts into structured musical outputs. It performs across styles including Chinese music, pop, R&B and soul, hip hop, rock, funk, electronic, classical, and jazz. Using world knowledge and music-domain reasoning, the model represents both the “grammar” of music.Starting Price: Free -
12
Apodex 1.1
Apodex
Apodex 1.1 is an online workbench for complex research and professional tasks that takes reasoning out of the report and puts it into the execution of real work. It is built to work through a task from start to finish using files, search, code execution, tool calling, and Agent Team coordination inside one long-running workflow. Users can bring in papers, datasets, spreadsheets, images, or code, while the model reads and manipulates files, writes and runs analysis code, checks intermediate results, revises its plan, and traces conclusions back to the underlying data and artifacts. Apodex 1.1 maintains task state across long chains of work, handles new feedback during execution, recovers from failures, and keeps the plan, steps, artifacts, dependencies, exceptions, and next actions visible. In Deep Discover mode, an asynchronous Agent Team breaks work into parallel subtasks, continuously feeds useful findings back into the main task.Starting Price: $25 per month -
13
Jev
TypeSafe AI
Jev is a System One Model from TypeSafe AI designed for fast, structured decision-making inside software and automated workflows. Unlike conventional large language models that generate free-form text token by token, Jev accepts unstructured state and returns predefined type-safe values with calibrated probabilities and confidence scores. The model uses a parallel sampling architecture and a training approach called Reinforcement Learning for Calibrated Decisions, which is optimized for producing reliable probabilistic decisions rather than conversational responses. Jev is intended for tasks such as classification, routing, scoring, extraction, branching logic, verification, guardrails, and other AI-powered workflows where structured outputs can plug directly into application code. TypeSafe reports end-to-end response times ranging from about 70 to 500 milliseconds and says Jev can be substantially faster and more efficient than frontier language models on System One-style workloads.Starting Price: Input: $0.042 / 1M tokens -
14
MiMo-V2.6-Pro-UltraSpeed
Xiaomi Technology
MiMo-V2.6-Pro-UltraSpeed is a high-speed serving mode for Xiaomi MiMo’s flagship MiMo-V2.6-Pro model, designed for latency-sensitive AI workloads. It delivers the same underlying model quality as MiMo-V2.6-Pro while providing output speeds of up to 20 times faster. The model supports coding, agentic automation, multimodal reasoning, visual design, research, and other complex tool-using workflows. Its capabilities include software engineering, frontend creation, presentation design, 3D modeling, interactive world generation, computer use, and multimodal analysis. MiMo-V2.6-Pro-UltraSpeed is intended for real-time applications where the capabilities of MiMo-V2.6-Pro are needed with substantially faster generation. It is available through MiMo Desktop and the Xiaomi MiMo API Platform.Starting Price: $4.35 per 1 million tokens inp -
15
Altar-1
Aikido Security
Aikido Altar is an open-weight security model built to bring frontier-grade defensive security intelligence into infrastructure organizations' control. It is designed for sovereign security environments where sensitive source code, architecture documentation, vulnerability findings, and other internal context cannot be sent to third-party inference services. Altar is based on GLM-5.3 and uses quantization and expert pruning to reduce the model from 1.51 TB at full precision to 328 GB while preserving most of the parent model’s reasoning and security capabilities. It retains 168 of the original 256 routed experts per backbone expert layer and uses a W4A16 representation, making deployment more practical for agentic security workloads with large and growing context windows. Expert selection was calibrated using internal pentesting traces and multilingual data, with no customer data involved, to preserve cybersecurity, coding, and language understanding.Starting Price: $350 per month -
16
Holo4
H Company
Holo4 is a family of generalist agentic AI models from H Company designed to operate software across graphical interfaces, code environments, MCP servers, APIs, desktop applications, the web, and mobile devices. The family includes a 27B dense model and a 35B-A3B Mixture-of-Experts model with 3B active parameters, both supporting a 256K context window. Unlike agents optimized around a single interface, Holo4 can click and type on screens, write and execute code, and call MCP or API tools depending on what a task requires. The models were trained with supervised fine-tuning and reinforcement learning on agentic workflows spanning desktop, web, MCP/API, mobile, coding, multimodal reasoning, and GUI grounding. Holo4 27B achieved 85.2% on OSWorld at a reported cost of $0.08 per task, while Holo4 35B-A3B scored 80.8% at $0.05 per task in H Company's evaluation.Starting Price: $0.40 per 1M tokens (input) -
17
Mercury Voice
Inception
Mercury is a family of diffusion large language models built to deliver frontier LLM quality at significantly higher speeds, running at 1,000+ tokens per second on commercial NVIDIA GPUs for instant, in-the-flow AI applications. The models are OpenAI-compatible and designed as drop-in replacements for traditional LLMs, making them easier to integrate into existing AI stacks. Mercury 2.5 is the family’s most intelligent reasoning dLLM, built for complex applications where both quality and speed matter. It supports a 260K context window, reasoning, tool use, and structured output, with use cases including rapid coding iteration, agents and subagents, customer support, and enterprise search. Mercury Voice is optimized for voice agents and delivers time-to-first-token under 170 ms while supporting reasoning, tool use, structured output, and a 128K context window. It is suited to applications such as customer support, patient care, education, and gaming.Starting Price: $0.04 per 1M tokens -
18
Mistral Medium 3.5
Mistral AI
Mistral Medium 3.5 is a flagship 128B dense model that merges instruction-following, reasoning, and coding into a single set of weights. It has a 256K context window, configurable reasoning effort, multimodal input, strong agentic capabilities, and is built for long-horizon tasks that require reliable multi-tool use and structured outputs. The model powers remote agents in Mistral Vibe, allowing coding sessions to run in the cloud, work through long tasks independently, and execute multiple sessions in parallel. Agents can be launched from the Vibe CLI or Le Chat, with local CLI sessions moved to the cloud while preserving history, task state, and approvals. Users can inspect file diffs, tool calls, progress states, and questions as work proceeds. Each coding session runs in an isolated sandbox and can handle module refactors, test generation, dependency upgrades, CI investigations, and bug fixes, then open a GitHub pull request for review.Starting Price: $14.99 per month -
19
Griffin
Tavus
Griffin is Tavus’s first Human Interaction Model, a full-duplex video-to-video system designed to understand and generate face-to-face human interaction in real time. Unlike traditional conversational systems that process speech recognition, language modeling, and speech synthesis sequentially, Griffin continuously perceives audio and video, decides when and how to respond, and generates expressive speech and video at the same time. It pays attention not only to words, but also to tone, gaze, facial expression, gestures, pauses, interruptions, and the surrounding visual context. The model can backchannel, yield or take the conversational turn, react to interruptions, remain engaged during silence, and adjust its emotional tone and nonverbal behavior as the interaction changes. Griffin generates the full visual scene from a reference image, including the face, body, hands, chair movement, shadows, and background, rather than animating only a talking head.Starting Price: $22 per month -
20
Mercury 3
Inception
Mercury 3 is a hosted generative AI model from Inception for developers building interactive applications. It uses a diffusion-based approach to generate text in parallel and supports streaming responses, tool use, structured outputs, and multimodal input including text, images, and audio. Through the Inception API, teams can use it for conversational assistants, voice applications, search and browsing workflows, tutoring, and other agentic tasks that benefit from responsive generation. The model is available through Inception’s API and named inference partners, with OpenAI-compatible API access for existing applications. Developers can evaluate output quality and latency, then integrate the model into their software using the documented API and SDK ecosystem. Usage is token-metered, with a free allocation and paid input/output token rates; enterprise options include dedicated capacity and support.Starting Price: $0.10/1M input tokens -
21
Abliterated Large V2
Abliteration.ai
Abliterated Large V2 is a large language model developed by Abliteration AI for advanced reasoning, cybersecurity research, AI red teaming, and autonomous agent workflows. Based on Z.ai's GLM 5.3 architecture, the model uses weight-level modifications designed to reduce refusal behavior while preserving coding and reasoning capabilities. It supports a one-million-token context window, enabling analysis of extensive source code repositories, technical documentation, and lengthy task histories. The model is optimized for security research, vulnerability analysis, exploit testing, and complex software engineering tasks. Its capabilities include tool calling, structured outputs, configurable reasoning, and integration with external applications through compatible APIs. Abliterated Large V2 is available as a hosted text-only model through Abliteration AI and supported third-party AI inference providers.Starting Price: $3 per 1M tokens (input) -
22
RoBERTa
Meta
RoBERTa builds on BERT’s language masking strategy, wherein the system learns to predict intentionally hidden sections of text within otherwise unannotated language examples. RoBERTa, which was implemented in PyTorch, modifies key hyperparameters in BERT, including removing BERT’s next-sentence pretraining objective, and training with much larger mini-batches and learning rates. This allows RoBERTa to improve on the masked language modeling objective compared with BERT and leads to better downstream task performance. We also explore training RoBERTa on an order of magnitude more data than BERT, for a longer amount of time. We used existing unannotated NLP datasets as well as CC-News, a novel set drawn from public news articles.Starting Price: Free -
23
ESMFold
Meta
ESMFold shows how AI can give us new tools to understand the natural world, much like the microscope, which enabled us to see into the world at an infinitesimal scale and opened up a whole new understanding of life. AI can help us understand the immense scope of natural diversity, and see biology in a new way. Much of AI research has focused on helping computers understand the world in a way similar to how humans do. The language of proteins is one that is beyond human comprehension and has eluded even the most powerful computational tools. AI has the potential to open up this language to our understanding. Studying AI in new domains such as biology can also give insight into artificial intelligence more broadly. Our work reveals connections across domains: large language models that are behind advances in machine translation, natural language understanding, speech recognition, and image generation are also able to learn deep information about biology.Starting Price: Free -
24
XLNet
XLNet
XLNet is a new unsupervised language representation learning method based on a novel generalized permutation language modeling objective. Additionally, XLNet employs Transformer-XL as the backbone model, exhibiting excellent performance for language tasks involving long context. Overall, XLNet achieves state-of-the-art (SOTA) results on various downstream language tasks including question answering, natural language inference, sentiment analysis, and document ranking.Starting Price: Free -
25
Hume AI
Hume AI
Our platform is developed in tandem with scientific innovations that reveal how people experience and express over 30 distinct emotions. Expressive understanding and communication is critical to the future of voice assistants, health tech, social networks, and much more. Applications of AI should be supported by collaborative, rigorous, and inclusive science. AI should be prevented from treating human emotion as a means to an end. The benefits of AI should be shared by people from diverse backgrounds. People affected by AI should have enough data to make decisions about its use. AI should be deployed only with the informed consent of the people whom it affects.Starting Price: $3/month -
26
FreedomGPT
Age of AI
FreedomGPT is a 100% uncensored and private AI chatbot launched by Age of AI, LLC. Our VC firm invests in startups that will define the age of Artificial Intelligence and we hold openness as core. We believe AI will dramatically improve the lives of everyone on this planet if it is deployed responsibly with individual freedom as paramount. It was created to showcase the inevitability and necessity of unbiased and censor free AI. Most importantly it is 100% private. If generative AI is going to be an extension of the human psyche it must not be involuntarily exposed to others. A central Age of AI investing thesis is that everyone and every organization will need their own private LLM. We strive to invest in companies that make this a reality across numerous industry verticals.Starting Price: Free -
27
CodeGen
Salesforce
CodeGen is an open-source model for program synthesis. Trained on TPU-v4. Competitive with OpenAI Codex.Starting Price: Free -
28
StarCoder
BigCode
StarCoder and StarCoderBase are Large Language Models for Code (Code LLMs) trained on permissively licensed data from GitHub, including from 80+ programming languages, Git commits, GitHub issues, and Jupyter notebooks. Similar to LLaMA, we trained a ~15B parameter model for 1 trillion tokens. We fine-tuned StarCoderBase model for 35B Python tokens, resulting in a new model that we call StarCoder. We found that StarCoderBase outperforms existing open Code LLMs on popular programming benchmarks and matches or surpasses closed models such as code-cushman-001 from OpenAI (the original Codex model that powered early versions of GitHub Copilot). With a context length of over 8,000 tokens, the StarCoder models can process more input than any other open LLM, enabling a wide range of interesting applications. For example, by prompting the StarCoder models with a series of dialogues, we enabled them to act as a technical assistant.Starting Price: Free -
29
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 -
30
Code Llama
Meta
Code Llama is a large language model (LLM) that can use text prompts to generate code. Code Llama is state-of-the-art for publicly available LLMs on code tasks, and has the potential to make workflows faster and more efficient for current developers and lower the barrier to entry for people who are learning to code. Code Llama has the potential to be used as a productivity and educational tool to help programmers write more robust, well-documented software. Code Llama is a state-of-the-art LLM capable of generating code, and natural language about code, from both code and natural language prompts. Code Llama is free for research and commercial use. Code Llama is built on top of Llama 2 and is available in three models: Code Llama, the foundational code model; Codel Llama - Python specialized for Python; and Code Llama - Instruct, which is fine-tuned for understanding natural language instructions.Starting Price: Free
