AI Providers

Find the right AI for your task

Explore the companies building AI, the platforms you can use, and the models behind them. This initial international directory covers selected major providers, open model developers and specialist services. It is not an exhaustive register of every company, model release or regional offering.

Sources checked: 14 October 2026. Model strengths below summarize provider documentation; they are not Syntropy Nexus benchmark results. Availability varies across chat apps, APIs, plans and regions. Current featured models appear below; official catalogs also identify older versions, previews and retirements.

Desktop download hub

Official sources checked: 15 October 2026. Choose one app or explore several. Links take you to the provider’s official download or installation page, where you select the latest compatible installer. Windows and macOS links may share the same operating-system selector.

Before you download: check the provider’s current device requirements, country availability and plan. A desktop app does not necessarily run its AI model offline. Browser extensions, terminal tools and model weights are different products.

For the remaining providers in this directory, a current first-party consumer desktop installer was not verified in this review. Use their linked official websites and documentation for web services, APIs, model downloads or integrations. This is a dated, selective directory; it is not a guarantee of every available app.

How this directory is organized

A provider builds or supplies AI. A platform is the app or service you use. A model is the system processing the task. One platform can offer several providers' models.

  • Large language models (LLMs): broad language, coding and reasoning capabilities, often with image or audio support.
  • Medium language models: an informal relative size description, without a universal parameter boundary. It does not mean a company is medium-sized.
  • Small language models (SLMs): models designed for a smaller resource footprint, sometimes suitable for local or edge deployment. Hardware, quantization, context length and task quality still matter.
  • Other model types: image and video generation, speech, embeddings, reranking, moderation, forecasting and physical/world models.

Many closed models do not disclose their parameter count. Names such as Mini, Flash or Haiku do not prove that a model belongs to a particular size class. Mixture-of-experts models also have different total and active parameter counts. Compare the actual task, quality, cost, latency, license and data controls.

A platform surface is where you interact with AI: a chat app, project workspace, coding tool, browser extension, office integration or developer console/API. Surfaces determine the workflow and tool access; the underlying model supplies the AI capabilities. The same model can appear in several surfaces. Surface descriptions below were checked on 15 October 2026.

General AI assistants and language model platforms

Anthropic brand mark

Anthropic Claude

Desktop apps: Claude: Windows downloads · macOS downloads. Chat and delegated work; choose the installer for your processor.

Anthropic develops the Claude model family and its assistant platform. Claude supports writing, document analysis, coding and work involving tools; Claude Code brings the models into software development workflows. Choose within the family according to the task's difficulty and response-speed needs.

Claude: where and how you work

  • Claude chat (web, desktop and mobile): conversational help with questions, writing and analysis. Open Claude · Official app downloads.
  • Projects: workspaces that group related conversations, reference material and instructions for ongoing work. A project supplies context; it is not a separate AI model. Projects explained.
  • Claude Code: an agentic software-development tool for working with codebases, editing files, running commands and tests. Its available interfaces include terminal, editor, desktop and web workflows. Claude Code overview and setup.
  • Claude Cowork: a way to delegate multi-step knowledge work across files and tools, such as preparing reports or organising material. Available surfaces and local-folder support differ. Cowork guide.
  • Claude in Chrome: the browser extension for reading webpages and carrying out browser tasks. It can work with Cowork or Claude Code. Download Claude in Chrome — official Chrome Web Store · Features, availability and permissions.
  • Claude developer platform/API: programmatic model access for building applications, distinct from the chat workspace. Developer documentation.

Which should you choose? Chat for a conversation; Projects for reusable context; Code for software work; Cowork for a delegated work deliverable; Chrome for work on webpages. These can work together. Check plan eligibility, administrator settings and requested permissions before installing or connecting tools.

Featured Claude models

  • Claude Fable 5.1: suited to demanding reasoning and long agent workflows; it prioritizes capability over response speed.
  • Claude Opus 5.5: suited to sustained coding and knowledge work, with a different capability and latency balance from Fable.
  • Claude Sonnet 5.5: balances intelligence and speed for frequent coding and general work.
  • Claude Haiku 5.5: emphasizes speed for high-volume classification, extraction and routing, rather than the family's most demanding reasoning tasks.

These are the current featured general models. Public parameter sizes are not established here. Older Claude generations and restricted-access research models should not be mistaken for the currently featured public lineup.

Official Claude model comparison and full documentation · Claude tutorials

OpenAI brand mark

OpenAI ChatGPT and Codex

Desktop apps: ChatGPT / Codex: Windows downloads · macOS downloads. The official download page includes ChatGPT with Work and Codex, plus ChatGPT Classic.

OpenAI develops GPT models and the ChatGPT assistant for conversation, writing, analysis and research. Codex focuses on software work. The API is a separate developer service; an API catalog entry does not guarantee the same model is selectable in every ChatGPT plan.

OpenAI: where and how you work

  • ChatGPT: the conversational application across web, desktop and mobile.
  • Projects: shared context for related work, including project files and instructions.
  • ChatGPT Work: a surface for delegated tasks using available tools and project context.
  • Codex: software-development work locally or through cloud tasks; the execution location affects access to your files and tools.
  • Developer API: model access inside your own applications, with separate setup and usage billing.

Official guide to ChatGPT Work, Projects and Codex. Features vary by plan and device.

Featured OpenAI models

  • GPT-6 Astra: the catalog's most capable choice for difficult reasoning and coding; use when task complexity justifies its greater resource demands.
  • GPT-6.1 Sol: offers near-Astra capability at lower cost, providing a useful balance for demanding everyday work.
  • GPT-6 Luna: targets focused, high-volume work where efficiency matters more than maximum reasoning capability.

OpenAI also offers specialized image, speech, embedding and moderation models. Use the official catalog for their current versions and the retirement record for older GPT releases. Model parameter sizes are not established here.

Official complete model catalog · ChatGPT getting started

Google brand mark

Google Gemini, AI Studio and Vertex AI

Desktop apps: Gemini: Windows downloads · macOS downloads. Windows 10+; Mac requires Apple silicon and macOS 15+. Features differ by operating system.

Google develops Gemini and supplies consumer, developer and cloud AI platforms. The Gemini assistant supports everyday multimodal work; Google AI Studio supports experimentation, while Vertex AI serves cloud deployment workflows.

Google: where and how you work

  • Gemini app: the conversational assistant for everyday multimodal tasks.
  • Google AI Studio: a developer workspace for testing prompts, exploring models and building prototypes with the Gemini API. AI Studio guide.
  • Vertex AI: Google Cloud's environment for developing and deploying AI applications. Explore Vertex AI.

The Gemini app, AI Studio and Vertex AI have different controls, access arrangements and billing. A model name alone does not identify the workspace you are using.

Featured Google models and services

  • Gemini 4 Argon: the current stable flagship for demanding reasoning and long-context work.
  • Gemini 3.8 Flash: designed for sustained agent and enterprise workflows.
  • Gemini 3.6 Flash: a previous fast multimodal general model.
  • Gemini 3.5 Flash-Lite: emphasizes economical, high-throughput processing.
  • Gemini 3.8 Live: low-latency voice interaction; Live Extended Thinking trades additional reasoning time for harder conversations.
  • Gemini 3.8 Flash TTS: expressive speech generation; Flash-Lite TTS emphasizes volume and efficiency.
  • Nano Banana 2.1: image generation and editing, including typography and iterative changes.
  • Gemini 3.5 Transcribe: speech transcription with speaker and timestamp capabilities.
  • Gemini Omni Flash: video generation in the catalog's preview category.

Official models, preview status and full catalog

Mistral AI brand mark

Mistral AI Le Chat and developer models

Mistral AI develops language, coding, vision and audio models. Its assistant Le Chat supports general work, while the developer platform supports deployment and specialized processing.

  • Mistral Large 4: current general multimodal model, with open weights.
  • Mistral Medium 3.5: focuses on agentic coding; its name is a provider designation, not a universal size boundary.
  • Mistral Small 4: combines instruction following, reasoning and coding.
  • Ministral 3 3B, 8B and 14B: smaller vision/text models with different memory and capability tradeoffs; 3B has the smallest listed footprint.
  • Codestral: specializes in code completion.
  • OCR 4.1: document extraction, including paragraph location and confidence information.
  • Voxtral Mini Transcribe 2: batch transcription; Voxtral Mini Transcribe Realtime targets live speech.
  • Voxtral TTS: generates speech; Voxtral Small supports audio instruction tasks.
  • Mistral Embed / Codestral Embed: retrieval representations for general text or code.
  • Shieldstral 1.0 / Mistral Moderation 2: safety classification rather than general chat.

Official full catalog, licenses and deprecated models

xAI brand mark

xAI Grok

Desktop apps: Grok Bot: Windows downloads · macOS downloads. A delegated-work desktop surface; check supported plans. This is distinct from Grok web chat.

The Grok ecosystem supplies conversational models and specialized media services. Grok 4.7 is the featured flagship for coding, tool use and configurable reasoning. Voice API supports spoken interaction, while Imagine API generates images and video. These services solve different tasks and should not be ranked by language-model size alone.

Official Grok model and service catalog

DeepSeek brand mark

DeepSeek chat and API

Desktop apps: DeepSeek Harness: Windows downloads · macOS downloads. A desktop agent for local files and complex tasks; Mac download requires Apple silicon and macOS 13+.

DeepSeek develops language models available through its chat service and developer API. The current API documentation identifies DeepSeek-V4.1-Flash, served by the deepseek-flash alias, and DeepSeek-V4-Pro. Flash is the efficiency-oriented choice; consult the provider's capability and pricing comparison before selecting Pro for a particular workload. Older accepted aliases may redirect and should not be treated as distinct current models.

Official current models and migration notices

Alibaba brand mark

Alibaba Qwen and Model Studio

Desktop apps: Qwen: Windows downloads · macOS downloads. Official Qwen desktop downloads; select your operating system on the provider page.

Alibaba's Qwen ecosystem covers language and multimodal AI. Model Studio hosts Qwen and third-party models; hosting a model does not make Alibaba its developer.

Featured text models include Qwen3.8-Max, Qwen3.7-Plus and Qwen3.8-Flash. Max and Plus also appear in image/video understanding; the catalog should guide task-specific comparisons rather than assumptions based only on names. Qwen3.8-Omni-Flash handles multimodal understanding; its Realtime variant supports live audio/video conversation. Qwen-Image-3.0-Pro generates images, Wan3.0-Video generates video, and the HappyOyster 1.0 family supports interactive world-model tasks. Dedicated Qwen Audio models provide speech generation and transcription.

Official recommended models and complete category catalogs

Moonshot AI brand mark

Moonshot AI Kimi

Desktop apps: Kimi: Windows downloads · macOS downloads. Kimi Work supports Windows and Apple silicon Macs; Intel Mac supports Chat only.

Moonshot AI develops Kimi, an assistant with web search, multimodal reasoning and long-context conversations, plus developer access.

  • Kimi K2.6: faster conversation and question answering.
  • Kimi K3: the assistant's most capable general choice for chat and agent tasks.
  • Kimi K3 Swarm: a mode for broad search and batch processing across coordinated agents; it is not simply a larger chat-model size class.

Official Kimi model and platform overview

Z.ai brand mark

Z.ai GLM

Z.ai supplies GLM models through its chat and developer ecosystem, including a coding plan. GLM-5.3 is the current flagship described for software engineering and agent tasks. GLM-5.3-Flash is a multimodal coding model focused on visual programming. Verify the exact model and endpoint offered by a plan before relying on an integration.

Official GLM models and developer overview

MiniMax brand mark

MiniMax agents, coding, video and audio

Desktop apps: MiniMax Design: Windows downloads · macOS downloads. A specialist creative studio; macOS 13+ or Windows 10+ x64.

MiniMax develops models and platforms spanning work agents and creative media.

  • MiniMax M3: multimodal coding with a one-million-token context window.
  • M3.1 Flash Preview: adds tunable thinking depth; currently limited to M Plan and MiniMax Code.
  • M2.7 / M2.7 Highspeed: older engineering/office models; Highspeed emphasizes faster inference. The family is scheduled for retirement on 18 November 2026, with access changes beginning 19 October.
  • MiniMax H3: multimodal video generation; H3 Max is a faster post-trained variant supplied with fal.ai.
  • Speech 2.8 HD: emphasizes speech quality; Speech 2.8 Turbo prioritizes speed.
  • Music 3.0: music generation, with important API access changes described in the catalog.

Official model lineup and retirement/access notices

Perplexity brand mark

Perplexity search and research

Desktop apps: Perplexity: Windows downloads · macOS downloads. Desktop research and Personal Computer; availability and agent features depend on plan and rollout.

Perplexity supplies a research-oriented assistant and developer services that combine models with search. Its Agent API offers tools, multi-step research and model selection across several developers. Sonar is Perplexity's own search-oriented model family. The official migration guide maps earlier Sonar offerings to Agent API presets: fast for quick/everyday research, low for multi-hop browsing and high for demanding deep research. A preset configures a workflow; it is not a language-model size class.

Official Sonar and Agent API comparison

AI21 brand mark

AI21 enterprise AI systems

AI21 focuses on enterprise AI systems. Its current offering includes an Intelligent Gateway for model routing, caching and context compaction, and Dedicated Model services for customized models. These are deployment and optimization services, so comparisons should consider the underlying selected model and the organization's workload rather than treating the gateway as a single chatbot model.

Official current AI21 platform

Cohere brand mark

Cohere Command and enterprise retrieval

Cohere develops models for enterprise language, search and document workflows. Command A+ combines vision, reasoning, translation and agent capabilities. Command A focuses on multilingual retrieval and tool use. Command R7B offers a smaller, faster option. Command A Translate, A Reasoning and A Vision specialize in translation, reasoning and image/document understanding respectively. North Small Translate targets translation across more than 50 languages, while North Mini Code targets agentic coding; their names do not establish a shared size category. Embed produces retrieval representations; Rerank orders search results; Parse extracts document data; Transcribe recognizes speech. Aya expands multilingual coverage.

Official current models, strengths and retirement status

Meta brand mark

Meta Meta AI and developer models

Desktop apps: Meta AI: macOS downloads. Official Mac app. A Windows desktop download was not verified in this review.

Meta develops consumer AI and model families. Meta AI is its assistant platform, while its developer ecosystem includes the Meta Model API and Muse Code. The current developer catalog features Muse Wonder, Muse Wonder Full, Muse Spark and Muse Glimmer, alongside Muse Image, Muse Voice Transcribe and SAM 3.1 for specialized image, speech and visual segmentation tasks. Muse Glimmer targets local agent workflows. Consult the exact model cards for differences, release status and licenses; the Llama catalog remains a separate historical and open-model resource.

Official Meta developer catalog and model cards

Microsoft brand mark

Microsoft Copilot

Desktop apps: Copilot: Windows downloads · macOS downloads. Use Microsoft’s download selector; personal and organizational features may differ.

Microsoft supplies the Copilot app and AI capabilities integrated into its productivity ecosystem. Its practical fit is assistance within familiar work tools, including conversation and content creation. Copilot is a platform, not one fixed model: the available features and underlying models depend on the product and plan. Microsoft's separately published model releases should be evaluated through their individual model cards rather than assumed to be identical to Copilot's current backend.

Microsoft: where and how you work

  • Copilot app: the general conversational assistant.
  • Microsoft 365 Copilot: assistance within Microsoft's productivity environment and supported work applications.
  • Copilot Studio: a graphical environment for building and managing agents, including extensions to Microsoft 365 Copilot. Official comparison and agent-building guide.

These are distinct products and workspaces; permissions, licences and available integrations differ.

Official Copilot getting-started video series

Open models, smaller footprints and enterprise deployment

IBM brand mark

IBM Granite and watsonx

IBM develops Granite models and offers watsonx for enterprise AI. The Granite 4.2 language lineup includes 3B, 8B and 30B variants: smaller variants reduce the weight footprint, while the 30B model targets greater capability at a larger resource cost. This is an editorial deployment tradeoff, not a guarantee of quality on your task.

Granite also includes Speech for transcription, Vision for documents/charts/images, Guardian for harmful-content detection, Embedding for retrieval and Time Series for forecasting. These specialist families serve different purposes and require their own evaluation.

Official Granite family and model selector

Ai2 brand mark

Ai2 Olmo and open research

The Allen Institute for AI develops open research models. Olmo 3 publishes weights, data and training resources so researchers can examine more of the development process.

  • Olmo 3 7B: the smaller language-model variant, with Base, Think and Instruct options.
  • Olmo 3 32B: a larger variant with the same broad release categories.
  • Base: a training foundation; Instruct: tuned to follow instructions; Think: tuned for reasoning. These describe behavior and training stage, not separate companies or platforms.

Official Olmo releases, research and access

TII brand mark

TII Falcon

The Technology Innovation Institute develops the Falcon family. Falcon Perception targets perception tasks. Falcon-H1 uses a hybrid Transformer/Mamba architecture for general language processing. Falcon-E emphasizes CPU and edge deployment. Falcon Arabic and Falcon-H1-Arabic target Arabic-language work. Falcon Mamba 7B explores a state-space language architecture. The catalog also preserves older Falcon releases; check the license for the exact model, as terms differ.

Official Falcon models and individual licenses

NVIDIA brand mark

NVIDIA Nemotron, Cosmos, NeMo and NIM

Desktop apps: ChatRTX: Windows downloads. Local AI demo requiring a supported NVIDIA RTX GPU. Check current hardware requirements; macOS download not verified.

NVIDIA supplies AI models and deployment infrastructure. Nemotron targets reasoning and agent applications; Cosmos targets physical and world-model applications. NeMo supports model and agent development, while NIM packages inference services. A NIM-hosted third-party model remains attributable to its original developer. Select the exact model card for capabilities, size and license before deployment.

Official NVIDIA AI model and platform overview

Amazon Web Services brand mark

Amazon Web Services Nova

Desktop apps: Amazon Quick: Windows downloads · macOS downloads. Workplace assistant; Windows x64 and Apple silicon Mac. Enterprise setup may require an administrator.

AWS develops Amazon Nova and supplies cloud AI services. Nova 2 Lite emphasizes efficient reasoning for everyday tasks; Nova 2 Sonic supports conversational speech-to-speech applications. Nova Act is a browser-workflow agent service powered by a customized model; Nova Forge supports building customized models. Services and models should be compared separately.

Official Nova models and services

Image, video and audio specialists

Adobe brand mark

Adobe Firefly

Desktop apps: Creative Cloud: Windows downloads · macOS downloads. Installs and manages creative desktop apps with AI features. Firefly’s web service is a separate surface.

Adobe supplies creative software and the Firefly platform for generating and editing images, video, audio and designs. Firefly Image 5 is an Adobe image-generation model; Firefly also hosts partner models from other developers. Its practical fit is creative ideation and iterative editing within a creative workflow. Model availability and commercial-use terms should be checked for the selected model, including partner models.

Official Firefly platform and model options

Black Forest Labs brand mark

Black Forest Labs FLUX

Black Forest Labs develops FLUX visual models and supplies a playground and API. FLUX 3 is the current family covering image/video generation and editing, synchronized audio and robotics applications. FLUX.2 remains supported for production image generation and editing; FLUX.1 Kontext is retained in its own documentation category. Choose according to the required media capability and deployment terms.

Official FLUX families and developer documentation

Runway brand mark

Runway creative media platform

Runway supplies a creative AI platform and API for video, images and related production workflows. Its catalog includes Gen-4.5, Gen-4 Turbo, Aleph 2.0, Act-Two and Gen-4 Image offerings, plus models from other developers. Video generation, video editing, performance transfer and image generation are different tasks; inspect the exact model's supported inputs and output settings. Hosted models such as Veo, Kling or Seedance should retain their original developer attribution.

Official Runway model index and supported capabilities

Stability AI brand mark

Stability AI creative generation

Stability AI develops generative media models and offers API, self-hosted and enterprise workflows. Stable Diffusion supports image generation, while Stable Audio 3.0 focuses on audio creation and experimentation. Its broader model catalog includes image, video, audio and 3D categories. Compare the exact version, licensing conditions and deployment route before adopting a model for production.

Official Stability AI models and deployment options

Providers that host and serve models

Together AI brand mark

Together AI

Together AI provides serverless, batch and dedicated inference, training and compute services. Its current catalog hosts models such as Kimi K3, MiniMax M3, Qwen and GLM. The platform is useful when an organization wants hosted access or customization for open models. Model strengths belong to the exact selected model and developer, while hosting performance depends on the deployment.

Official Together AI platform and catalog

Fireworks AI brand mark

Fireworks AI

Fireworks supplies infrastructure for deploying and specializing AI models. Its platform emphasizes inference and model customization for production applications. Evaluate the selected model, hosting configuration and data controls separately; a hosting service is not a new language-model size class.

Official Fireworks platform

Groq brand mark

Groq

Groq supplies inference infrastructure designed for fast model execution. Its platform is a route for serving supported models, rather than a single general-purpose assistant model. Check the live supported-model list and test latency and quality with your own workload.

Official Groq platform

How to choose

Start with the task: writing, coding, research, document processing, image/video creation, speech or local deployment. Then compare quality on representative examples, privacy controls, license, total cost, response time and integration effort. Keep an untouched test set, include failures and avoid assuming that a newer or larger model always performs better.

Coverage note: this is a dated first edition. Additional regional developers, specialist providers, open-weight variants, legacy versions and deployment platforms remain to be cataloged. Linked official catalogs provide the broader current lineup; this page does not claim to list every AI provider in the world.

Provider logos and site icons identify the companies described. They remain their owners’ trademarks and do not imply affiliation, sponsorship or endorsement of SNAF.