AI models and what they're good for
The models behind today's AI apps, in one line each: who makes it, what it's good for, and where you can use it.
Chat and reasoning models
-
GPT-6 Astra
OpenAI
OpenAI's top model for hard coding, research and long multi-step work.
-
GPT-6.1 Sol
OpenAI
Codes and operates apps nearly as well as Astra for about a fifth the API price.
-
GPT-6 Luna
OpenAI
Cheap, fast model for summaries, data extraction and quick answers.
-
Claude Fable 5.1
Anthropic
Slowest, priciest Claude, for demanding reasoning and long agent work when Opus falls short.
-
Claude Opus 5.5
Anthropic
Anthropic's default pick for coding agents and long knowledge work; costs less than Fable.
-
Claude Sonnet 5.5
Anthropic
Faster, cheaper Claude for everyday work; Anthropic reports scores close to Opus 5.5.
-
Claude Haiku 4.5
Anthropic
Cheapest, fastest Claude for simple, high-volume tasks.
-
Gemini 4 Argon
Google
Long reasoning for code, legal and finance work. Limited access so far.
-
Gemini 3.1 Pro
Google
Gemini app's Pro model for reasoning, coding and long documents.
-
Gemini 3.8 Flash
Google
Low-cost, fast model for coding agents and multi-step work.
-
Grok 4.7
SpaceXAI (formerly xAI)
Coding and long, multi-step professional tasks via API and coding tools.
-
Muse Spark 1.3
Meta
Meta's model for agent tasks and coding; powers Meta AI and Meta Muse.
-
Mistral Large 4
Mistral AI
Mistral's largest model, for coding, agents and document work. Preview only.
-
Qwen3.8-Max
Alibaba
Alibaba's top hosted model for coding and office-style tasks.
-
DeepSeek-V4-Pro
DeepSeek Open weights
DeepSeek's largest model: coding agents, long tool tasks, 1M-token context, low price.
Open-weight language models
-
gpt-oss (20B, 120B)
OpenAI Open weights
Reasoning and tool use on your own computer; 20B runs with 16 GB of memory.
-
Qwen3.8 (plus Qwen3.6 and Qwen3.5)
Alibaba Open weights
Local coding, agents and questions about images; a huge range of model sizes.
-
Gemma 4
Google Open weights
Fast everyday chat, writing and questions about images on laptops and phones.
-
Llama 3.3 and Llama 4
Meta Open weights
General chat; widely supported, but no new Llama since April 2025.
-
Muse Glimmer
Meta Open weights
Local agents and tool use on a 32 GB Mac or single GPU.
-
DeepSeek-V4.1-Flash and R1 distills
DeepSeek Open weights
Low-cost reasoning and coding via app or API; smaller models trained on R1 run locally.
-
Mistral Small 4 and Mistral Medium 3.5
Mistral AI Open weights
European option for chat, coding agents and documents; reads images.
-
Kimi K3
Moonshot AI Open weights
Long coding sessions and agent work; open weights need a GPU cluster.
-
GLM-5.3
Z.ai (formerly Zhipu AI) Open weights
Long-running coding agents and security work at low API prices.
-
MiniMax-M3
MiniMax Open weights
Coding and agents with 1M-token context; reads images and video.
-
Phi-4 family
Microsoft Open weights
Small models for math, reasoning and running on modest hardware.
-
Nemotron 3 and Nemotron 3.5
NVIDIA Open weights
Agents and long documents; small Nano models run fast locally.
-
MiMo-V2.6
Xiaomi Open weights
Coding and agents; handles text, images, audio and video.
-
Inkling
Thinking Machines Lab Open weights
A general base for companies to fine-tune; reads text, images and audio.
Image models
-
GPT Image 2.5 (Flare and Sunburst)
OpenAI
All-round image making; accurate text in images; edits that leave the rest untouched.
-
Nano Banana 2.1
Google
Fast, cheap image making and chat-style editing; infographics; keeping characters consistent.
-
Nano Banana Pro (Gemini 3 Pro Image)
Google
Detailed 4K images, complex layouts, diagrams and mockups with accurate text.
-
Midjourney V8 (V8.1 and V8.2)
Midjourney
Artistic, polished images with a strong look; style references and personal taste profiles.
-
FLUX 3 Image
Black Forest Labs
Placing objects exactly where you draw boxes; repeated edits that leave the rest unchanged.
-
FLUX.2
Black Forest Labs Open weights
Images and edits with up to 10 references; open versions to run locally.
-
FLUX.1 ([schnell], [dev], Kontext)
Black Forest Labs Open weights
Local image making with thousands of community add-ons; [schnell] is fast and allows commercial use.
-
Stable Diffusion 3.5 and SDXL
Stability AI Open weights
Running on modest hardware; huge range of community styles and add-ons, mostly SDXL.
-
Qwen-Image (3.0, 2.1, 2512)
Alibaba Open weights
Text-heavy posters, infographics and page layouts; open versions run locally.
-
Seedream 5.0 (Lite and Pro)
ByteDance
Product shots and marketing graphics; point-and-drag edits; text in many languages.
-
Ideogram 4.0
Ideogram Open weights
Logos, posters and typography; placing text and objects exactly where you want them.
-
Recraft V4.1
Recraft
Editable vector (SVG) graphics, icons and illustrations; print-ready design assets.
-
MAI-Image-2.6
Microsoft
Microsoft's image model for generating pictures and edits that combine several photos.
-
Grok Imagine Image 2.0
SpaceXAI (formerly xAI)
Images and precise edits inside Grok; combining up to five source images.
-
Z-Image-Turbo
Alibaba Open weights
Fast, realistic images on a home computer; English and Chinese text; free license.
-
Muse Image
Meta
Making and editing images in Meta AI, including combining several reference photos.
Video models
-
Veo 3.1
Google
Short, cinematic clips with sound; control first and last frames.
-
Gemini Omni 1.1 Flash
Google
Make and edit video by chatting; mix text, image, video, audio inputs.
-
Gen-4.5
Runway
Text- or image-to-video clips up to 10 seconds, inside Runway's editing tools.
-
Kling 3.0
Kuaishou (Kling AI)
Clips up to 15 seconds with native audio and lip-synced dialogue.
-
Seedance 2.5
ByteDance
Long single shots, up to 30 seconds, steered by many reference files.
-
MiniMax H3
MiniMax Open weights
Ad and product videos with stereo sound; open weights for self-hosting.
-
Wan 3.0
Alibaba
Clips up to 30 seconds at 1080p from text, images, audio or documents.
-
Wan 2.2
Alibaba Open weights
Free open model for short clips on your own GPU.
-
Ray3.2
Luma AI
Directing a shot frame by frame with keyframes; HDR for post-production.
-
FLUX 3 Video
Black Forest Labs
Clips up to 20 seconds with dialogue and sound effects, via API.
-
Grok Imagine Video 1.5
SpaceXAI (formerly xAI)
Quick, low-cost clips up to 15 seconds; cheaper Lite version too.
-
LTX-2.5
Lightricks Open weights
Local video with synced audio; free if your yearly revenue is under $10M.
-
Muse Video
Meta
Turning text prompts into video clips with built-in sound.
Speech and music models
-
Eleven v4
ElevenLabs
Expressive voiceovers, audiobooks and character dialogue; v4 Turbo for live voice agents.
-
Scribe v2
ElevenLabs
Transcripts with speaker labels and timestamps; Realtime version for live captions.
-
GPT-Live
OpenAI
Back-and-forth voice chat; listens and talks at the same time.
-
GPT-Transcribe and GPT-Live-Transcribe
OpenAI
Transcription of audio files; Live version for real-time captions.
-
gpt-4o-mini-tts
OpenAI
Cheap app voiceovers; steer tone, accent and pace with plain instructions.
-
Whisper (large-v3 and turbo)
OpenAI Open weights
Free, private transcription on your own computer in many languages.
-
Gemini 3.8 Flash TTS
Google
Expressive narration and two-voice dialogue; design a voice from a text description.
-
Gemini 3.8 Live
Google
Low-lag spoken conversations for voice agents; can see video and call tools.
-
Parakeet TDT 0.6B v3
NVIDIA Open weights
Very fast local transcription with timestamps; English and 24 other European languages.
-
Chatterbox
Resemble AI Open weights
Free voice cloning from a short clip; 23 languages; fine for commercial use.
-
Kokoro-82M
hexgrad Open weights
Tiny, fast text-to-speech for local apps; 54 voices in 8 languages.
-
Voxtral (Transcribe 2, Realtime, TTS)
Mistral AI Open weights
Low-cost transcription with speaker labels; open live transcription and voice cloning.
-
Suno v6
Suno
Full songs with vocals from a prompt; edit one section in plain language.
-
Lyria 3.5
Google
Full songs with vocals up to 3 minutes, from a text prompt or photo.
-
Eleven Music v2.5
ElevenLabs
Songs with vocals or instrumentals from a text prompt, editable section by section.
-
MAI-Transcribe-2 and MAI-Voice-2.1
Microsoft
Low-cost transcripts in 60 languages; text-to-speech with voice cloning in 23 languages.
"Open weights" means the model files can be downloaded and run on your own hardware, under the maker's license. Model names and versions change quickly; each name links to the maker's own page. Looking for apps instead? See the AI tools directory.