AI models and tools in 2026: major AI services and models in one guide
AI models and tools in 2026: major AI services and models in one guide

Category: Tutorials and guides | Published: September 16, 2026
Topics: Large language models (LLMs), multimodal AI, coding agents, local inference, and AI services
Author: Zoran Knežević | Reading time: About 15 minutes

The AI market can no longer be described by ChatGPT, Claude, and Gemini alone. Major companies are building families of models for different kinds of work:

  • Reasoning and complex problem solving
  • Everyday conversation and chat assistance
  • Software development and maintenance
  • Autonomous AI agents and multi-step workflows
  • Working with large documents and codebases
  • Web search and information synthesis
  • Image and video analysis
  • High-resolution image generation
  • Video generation with audio
  • Real-time voice conversations
  • Text-to-speech (TTS) and speech-to-text transcription
  • Music and sound-effect generation
  • Semantic embeddings and retrieval-augmented generation (RAG) knowledge bases
  • Local AI systems that run on your own hardware

This guide is organized primarily by AI companies, their models, and the work those models are designed to do. It then points to services where you can use them in practice.

The focus is on capable, current, operationally relevant models in 2026.

OpenAI

Main services:

OpenAI has several model families for general AI, coding, images, voice, transcription, and specialized tasks.

GPT-6 Astra

  • Details: GPT-6 Astra documentation
  • OpenAI’s most capable general-purpose model at the time of writing.
  • Designed for demanding work such as complex reasoning, software engineering, large codebases, computer use, scientific research, complex documents, professional knowledge work, and multi-step agent tasks.
  • Supports several reasoning levels and a context window of up to about 1.05 million tokens.

GPT-5.6 Sol

  • Details: GPT-5.6 Sol
  • The flagship model in the GPT-5.6 generation.
  • Suited to coding, research, professional work, science, analysis, and complex agents.
  • Supports a context window of about 1.05 million tokens and a large maximum output.

GPT-5.6 Terra

  • A mid-range model in the GPT-5.6 family.
  • Designed to balance intelligence, response speed, and usage cost.
  • A good fit for production applications that need more capability than Luna but do not require Sol for every request.

GPT-5.6 Luna

  • Details: GPT-5.6 Luna
  • The fastest and least expensive member of the GPT-5.6 family.
  • Well suited to classification, quick data extraction, short conversations, high-volume request processing, automation, and simpler agents.
  • It still supports reasoning and a large context window.

GPT Image 2.5 Sunburst

  • Details: OpenAI models
  • OpenAI’s most advanced line for image generation and editing.
  • Sunburst is tuned for high photorealism and visual quality.

GPT Image 2.5 Flare

  • A faster variant of the GPT Image 2.5 family.
  • Intended for everyday image generation where a strong balance of quality, speed, and cost matters.

GPT Image 2

  • Details: GPT Image 2
  • A model for text-to-image, image-to-image, image editing, and working with reference images.
  • Accepts combined text and image input.

GPT-Live 1

  • An OpenAI model for natural, real-time voice conversations.
  • Use it when an AI needs to listen actively, answer aloud, handle user interruptions, and maintain a live conversation.

GPT-Realtime 2.1 and 2.1 Mini

  • Reasoning models for real-time applications with tool use.
  • GPT-Realtime 2.1 Mini is a faster, more economical option for interactive applications.
  • Previous generation: GPT-Realtime 2.

Voice and audio models

  • GPT-Realtime-Translate: Streaming speech-to-speech translation for near real-time spoken translation.
  • GPT-Live-Transcribe: A model for very fast real-time transcription.
  • GPT-Realtime-Whisper: Streaming speech recognition based on the Whisper architecture.
  • GPT-Transcribe: The current high-accuracy speech-to-text model (GPT-Transcribe details). It supports audio files, streaming, multiple languages, keyword hints, and automatic language switching.
  • GPT-4o Mini TTS: A model that converts text into natural-sounding speech.
  • Whisper: A proven open speech-recognition model (Whisper details) for transcription, language identification, and translation.

Specialized OpenAI models

  • GPT-5.6 Cyber: A model for authorized cybersecurity research and security testing.
  • OpenAI Daybreak: A family of advanced cybersecurity models.
  • GPT-Rosalind: A specialized model for biology and life-sciences research.

Google DeepMind and Gemini

Main services:

Google has one of the broadest and most capable AI portfolios, with separate models for reasoning, coding, real-time audio, translation, transcription, images, video, embeddings, and robotics.

Gemini 3.8 Flash

  • Details: Gemini 3.8 Flash
  • The latest leading Flash model at the time of writing.
  • Designed for software engineering, autonomous agents, complex business processes, tool use, long context, and multimodal tasks.
  • Supports a context window of up to one million tokens with adjustable reasoning effort.

Gemini 3.7 Flash and 3.6 Flash

  • Gemini 3.7 Flash: A capable Flash generation focused on speed, reasoning, agents, and coding.
  • Gemini 3.6 Flash: A model optimized for code generation, spatial reasoning, and fast multi-step tasks (Gemini 3.6 Flash details).

Gemini 3.5 Flash and Flash-Lite

  • Gemini 3.5 Flash: A frontier model for agents, coding loops, and a one-million-token context window (Gemini 3.5 Flash details).
  • Gemini 3.5 Flash-Lite: A lightweight, low-cost option for processing large batches of requests, extraction, and low-latency systems.

Gemini 3.1 Pro and 2.5 Pro

  • Gemini 3.1 Pro: One of Google’s most capable models for deep reasoning (Gemini 3.1 Pro details). It supports text, images, video, audio, and PDFs with up to one million tokens of context.
  • Gemini 3 Flash: The previous generation, still common in stable integrations.
  • Gemini 2.5 Pro: A proven line for mathematics, STEM, complex coding, and multimodal analysis (Gemini 2.5 Pro details), with a context window of 1,048,576 tokens.

Google image models: Nano Banana and Imagen

Google’s current image models use the Nano Banana name:

  • Nano Banana Pro (gemini-3-pro-image): The most capable option for image creation and editing, complex compositions, typography, and reference images.
  • Nano Banana 2 (gemini-3.1-flash-image): A fast model for generation and interactive editing (Gemini 3.1 Flash Image), with resolutions up to 4K and search grounding.
  • Nano Banana 2 Lite (gemini-3.1-flash-lite-image): A lower-cost option for bulk image generation.
  • Imagen 4: The previous generation (Imagen details); its endpoints have been replaced by newer Gemini image models.

Google video models: Gemini Omni Flash and Veo 3.1

  • Gemini Omni Flash: A new generation of multimodal video models (Google video models). It supports text-to-video, image-to-video, audio input, video input, conversational editing, and character consistency.
  • Veo 3.1: A high-end video production model with built-in stereo audio generation (Veo 3.1 details). It supports 720p, 1080p, and 4K, horizontal and vertical formats, first- and last-frame control, and video extension.
  • Veo 3.1 Fast and Veo 3.1 Lite: Faster, optimized variants for producing large numbers of clips.

Google real-time voice, audio, and transcription

  • Gemini 3.8 Live: The current model for direct voice interaction, designed for voice agents with native audio input and output.
  • Gemini 3.8 Live Extended Thinking: Adds a reasoning step before answering for more complex voice assistants.
  • Gemini 3.5 Live Translate: A specialized model for simultaneous real-time speech translation.
  • Gemini 3.1 Flash TTS and 2.5 Pro TTS: Models for natural-sounding speech synthesis.
  • Gemini 3.5 Transcribe and Transcribe Live: Options for standard and streaming audio-to-text transcription.

Google embeddings and robotics

  • Gemini Embedding 2: The current generation for semantic search, RAG systems, classification, and clustering.
  • Gemini Robotics ER 2: An embodied reasoning model for understanding physical environments, analyzing video, spatial reasoning, and coordinating robots.
  • Gemini Robotics ER 1.6: The previous stable version for robotics applications.

Google Gemma open models

For your own servers and local setups, see Gemma:

  • Gemma 4: The latest open generation, with E2B, E4B, 12B, 26B A4B, 31B, and 12B Unified variants. It supports multimodal input and can run on laptops or mobile devices.
  • Gemma 3 and Gemma 3n: Popular variants with 270M, 1B, 4B, 12B, and 27B parameters, plus energy-efficient E2B and E4B versions.
  • TranslateGemma: 4B, 12B, and 27B models specialized for translation.
  • MedGemma, FunctionGemma, and EmbeddingGemma: Models focused on medicine, function calling, and local vector databases.

Anthropic Claude

Main services:

In 2026, Anthropic offers clearly differentiated levels of capability and speed:

  • Claude Fable 5.1: A top model for coding, long-running autonomous agents, research, science, and large projects (Claude Fable details).
  • Claude Mythos 5.1: A specialized version for cybersecurity research and life sciences (Claude Mythos details).
  • Claude Opus 5: A highly capable model for engineering, debugging, computer use, and business analysis (Claude Opus 5 details).
  • Claude Sonnet 5: A strong choice for coding, terminal and browser use, and multi-step agent tasks (Claude Sonnet 5 details).
  • Claude Opus 4.8 and Sonnet 4.6: Earlier high-capability versions that remain in active production use.
  • Claude Haiku 4.5: The fastest model for high-throughput requests, classification, and quick answers.

xAI and Grok

Main services:

  • Grok portal

  • xAI API console

  • Grok 4.6: The current flagship model supports text and images, a 500K context window, web and X search, and automatic code execution (Grok 4.6 details).

  • Grok 4.5: Specialized in software engineering, agents, science, and mathematics (Grok 4.5 details).

  • Grok 4.20 and 4.20 Multi-Agent: Models with a one-million-token context window and the ability to run several agents in parallel (Grok 4.20 details).

  • Grok 4.3: A fast model with adjustable reasoning effort (Grok 4.3 details).

  • Grok Imagine Image 2.0 and Imagine Video 1.5: Tools for generating photorealistic images and video clips.

  • Grok Voice API: A real-time platform for voice agents.

DeepSeek

Main services:

  • DeepSeek Chat

  • DeepSeek Platform API

  • DeepSeek V4 Pro (deepseek-v4-pro): A leading model for deep reasoning, advanced coding, agents, and long documents.

  • DeepSeek V4.1 Flash (deepseek-v4.1-flash): A fast version that combines thinking and non-thinking modes.

  • Compatibility: Supports OpenAI-compatible and Anthropic-compatible request formats, with very affordable token pricing.

Alibaba Qwen

Main services:

Qwen is a large ecosystem of models built with both mixture-of-experts (MoE) and dense architectures:

  • Qwen3.8-Max: The main model, with 2.4 trillion total parameters, a one-million-token context window, and advanced reasoning.
  • Qwen3.8 Flash: A fast option for agents, web search, and video processing.
  • Qwen3.8 27B: A strong open dense model (Qwen3.8 27B details).
  • Qwen3.7 Plus and Flash: Balance cost and strong capability for coding and text.
  • Qwen3 Coder Plus and Flash: Models specialized in software engineering.
  • Qwen3-VL Plus and Flash: Leading vision-language models for image, table, OCR, and video analysis.
  • Qwen3.5 Omni: Models that support simultaneous multimodal input.

Z.ai and GLM

Main service:

  • Z.ai Chat

  • GLM-5.3: An open-weight flagship model with hundreds of billions of parameters for coding, long-running agents, and tool use.

  • GLM-5.3 Flash: A more efficient version with multimodal input and low processing costs.

  • Previous versions: GLM-5.2, GLM-5.1, and GLM-5 remain widely available through platforms such as Mistral and Alibaba Studio.

Moonshot AI and Kimi

Main service:

  • Kimi portal

  • Kimi K3: Moonshot AI’s current leading model, known for working with very large documents, deep research, and automated presentation creation.

MiniMax

Main services:

MiniMax offers a full range of generative AI:

  • MiniMax M3: A model for coding, computer use, and a one-million-token context window (MiniMax M3 details).
  • MiniMax H3: An omni model that can generate video with native stereo audio.
  • MiniMax M2.7 and M2.5 Highspeed: Fast models for office work and coding.
  • MiniMax Speech 2.8 HD and Turbo: High-end speech synthesis.
  • MiniMax Music 3.0: Advanced song generation with lyrics (MiniMax Music 3.0 details).
  • MiniMax Hailuo 2.3 and Hailuo 02: High-definition video production at 1080p.

Meta AI and Llama

Services:

  • Meta AI

  • Meta AI Developer

  • Llama models

  • Muse Spark 1.3: A new Meta agent model for GitHub and coding.

  • Muse Glimmer: A model for local agents.

  • Muse Image and Muse Voice Transcribe: Tools for image and audio tasks.

  • Llama 4 Maverick: A leading multimodal model with 400B total parameters, 17B active parameters, and a one-million-token context window.

  • Llama 4 Scout: An efficient variant with a context window of up to 10 million tokens.

  • Llama 3.3 70B and Llama 3.2 Vision and Mini: Widely used local models for servers and mobile devices.

Mistral AI

Main services:

  • Mistral Le Chat

  • Mistral AI console

  • Mistral Medium 3.5: A frontier model with a 256K context window, tuned for structured output and agents (Mistral Medium 3.5 details).

  • Mistral Small 4: A hybrid model for reasoning and coding.

  • Ministral 3 (3B, 8B, 14B): Models suited to local hardware and edge devices.

  • OCR 4.1: A capable model for tables, handwriting, and PDF extraction.

  • Voxtral Mini Transcribe 2: A precise speech-to-text model.

Cohere

Main service:

  • Cohere

  • Command A+: An MoE model for translation into 48 languages, agents, and tool use (Command A+ details).

  • Command A: A solution for enterprise RAG systems (Command A details).

  • Aya Expanse 32B and Tiny Aya: Multilingual models for global use (Aya Expanse details).

  • Rerank 4 Pro: An industry-standard option for ranking results in RAG databases.

  • Parse 5: Extracts data from complex business forms and tables.

Microsoft AI and Phi

Services:

Amazon Nova and IBM Granite

Amazon Nova:

  • Amazon Nova and Amazon Bedrock
  • Nova 2 Lite: A one-million-token context window, multimodal input, agents, and MCP support.
  • Nova 2 Sonic: A speech model for conversational assistants.
  • Nova Multimodal Embeddings: A dedicated vectorization model for all media types.

IBM Granite:

  • IBM Granite
  • Granite 4.0 family (H Small, H Tiny, Micro, 1B, 350M): Open models under the Apache 2.0 license that combine Transformer and Mamba-2 architectures for secure enterprise deployment.

Multi-model services

These platforms provide access to models from several companies through one account or API:

  • OpenRouter - One API for hundreds of models, with automatic routing and price comparisons.
  • Together AI - Fast cloud inference for leading open models.
  • NVIDIA NIM - Optimized microservices for enterprise deployments.
  • Arena - Anonymous, real-time model comparisons from LMSYS.
  • Lumo - A private AI assistant from Proton.
  • Ecosia AI - An environmentally conscious AI chat service.
  • Public AI - Access to open models without commercial barriers.
  • Perplexity - A leading search engine that cites sources.
  • Google AI Mode - Google Search with an advanced generative layer.
  • Brave Ask - Search focused on user privacy.
  • Exa - An AI-native semantic search engine for software agents.
  • Vane - An open local AI search engine.

AI for research and document analysis

AI coding tools

  • Cursor - A next-generation AI editor for working across an entire project.
  • Claude Code - Anthropic’s agent for terminals and complex repositories.
  • OpenAI Codex - OpenAI’s specialized software agent.
  • GitHub Copilot - An integrated assistant for VS Code and JetBrains.
  • Gemini CLI - Google’s terminal-based coding partner.
  • Windsurf - An AI development environment with advanced context handling.
  • Cline - A popular open source agent for VS Code.
  • Aider - A terminal-based pair programmer for Git repositories.
  • OpenHands - An autonomous open source software engineer.
  • Replit, v0, Lovable, and Bolt - Platforms for quickly turning ideas into working web apps.
  • MiniMax Code - A coding agent optimized for MiniMax M3.

General-purpose AI agents

  • Manus - An autonomous agent for search, documents, and multi-step tasks.
  • Genspark - Agentic search with interactive Spark page creation.
  • Hermes Agent - An open agent from Nous Research.
  • Goose - A flexible local agent from Block.
  • OpenClaw - An advanced platform for building personalized agents.

Presentations and design

  • Gamma - Automatically create modern presentations and web documents.
  • Canva - An all-in-one visual design suite with built-in AI tools.
  • Adobe Firefly - Generative AI for creative work, designed for commercial use.

Image generation

  • Midjourney - A leading service for photorealistic and artistic images.
  • Ideogram - Strong at rendering accurate text and typography in images.
  • Recraft - Create vector graphics, icons, and 3D elements for design projects.
  • Leonardo AI - A complete creative studio with style controls.
  • Reve and Dreamina - Next-generation image generation and editing.
  • Bing Image Creator - A fast image generator with a free option.
  • Mage and Perchance - Web interfaces for diffusion models.

Generate images locally on your computer

  • ComfyUI - A professional node-based system for full workflow control.
  • Stability Matrix - A manager that installs popular interfaces with one click.
  • Invoke - A clean, professional environment for creative work.
  • Fooocus and Forge - Optimized local tools that are easy to use.
  • SD.Next and DiffusionBee - Useful options for PC and macOS users.
  • ComfyUI-ZLUDA - An optimization for AMD graphics cards.
  • Civitai - A large global community for downloading models, LoRAs, and styles.

AI video production

  • Google Flow - Google’s creative platform for Veo and Gemini models.
  • Runway - A high-end tool for filmmaking, rotoscoping, and video generation.
  • Kling AI - Known for realistic physics and motion in a frame.
  • Luma Dream Machine and Pika - Quickly turn ideas into animations.
  • Wan - Alibaba’s advanced video generator.
  • PixVerse and Genmo - Web platforms for dynamic effects.
  • FramePack Studio - Run video generation on a local computer.

Avatars and presenters

  • HeyGen - Create lifelike digital avatars with video translation and lip sync.
  • Synthesia - A corporate standard for training and educational videos.

AI music and composition

  • Suno - Generate complete songs with vocals and instruments across many genres.
  • MiniMax Music - An advanced open model for professional music creation.
  • Stable Audio - High-fidelity sound effects and music beds.
  • MusicGen - Meta’s open source audio-generation model.
  • AIVA, Mubert, SOUNDRAW, and Boomy - Tools for quickly producing music for creators.
  • ACE-Step and MusicGPT - Newer open solutions.

Text-to-speech, voice cloning, and audio separation

Local LLM systems: run AI on your computer

If you want more privacy and the ability to work without an internet connection:

  • Ollama - A simple command-line tool for running models such as Llama, Gemma, and Mistral.
  • LM Studio - A popular desktop app with a graphical interface and built-in API server.
  • Open WebUI - A ChatGPT-like interface that connects to Ollama models.
  • Jan and AnythingLLM - Useful apps for searching local documents without sending them elsewhere.
  • Msty, Cherry Studio, Chatbox, and LibreChat - Apps for working with local models and cloud providers side by side.
  • KoboldCpp and llama.cpp - Foundational tools for the local LLM ecosystem that can run quantized GGUF models.

Hardware calculators: which models can your hardware run?

Compare API prices and performance

Official benchmarks and leaderboards

Learning, prompt engineering, and AI tool directories

Conclusion

In 2026, successful AI use depends on choosing the right tool for the task:

  • For deep reasoning and complex coding: Gemini 3.8 Flash, Claude Sonnet 5, GPT-6 Astra, and DeepSeek V4 Pro.
  • For fast work on your own computer: Ollama or LM Studio with Gemma 4, Llama 4, or Ministral.
  • For visual and multimedia work: Midjourney, Ideogram, Veo 3.1, HeyGen, and Suno.

Save this guide to your bookmarks as a reference map of the current AI landscape.