LayersLayer 2 of 15
Models
Models are the trained systems that do the work: language models, and models for speech, images, video, code, translation and forecasting. This layer covers both closed models reached through an API and open models you can download.
The categories
24 categories and 649 products in Models.
CategoryWhat it coversProducts
3D generation & motion3D generation & motion covers Text/image-to-3D asset, scene, CAD and motion-capture generation models and APIs.25
Avatar / digital humansAvatar / digital humans covers talking-head, presenter and interactive real-time avatar video models and APIs.30
Background removal & mattingBackground removal & matting covers foreground segmentation, background removal and video matting models/APIs.26
Benchmarks & leaderboardsBenchmarks & leaderboards covers public benchmark suites and ranking sites that compare models, agents, retrievers and providers.34
Embedding modelsEmbedding models covers models that turn text, images or audio into numbers a search system can compare, used for search and retrieval.32
Frontier model APIsFrontier model APIs covers the most capable closed language models, reached through their makers' APIs.28
Image generation & editingImage generation & editing covers text/image-to-image generation and prompt-based editing models.36
Image upscaling & restorationImage upscaling & restoration covers super-resolution, face restoration, denoise and photo enhancement models/APIs.23
Lip-sync & dubbingLip-sync & dubbing covers video lip-sync models and AI dubbing/voice translation APIs.20
Machine translation APIsMachine translation APIs covers hosted APIs and models that translate text or documents between languages (NMT and LLM-based), usually billed per character.35
Model hubs / registriesModel hubs / registries covers the sites where model files and their documentation are published and downloaded.24
Multilingual / regional-language modelsMultilingual / regional-language models covers models built for specific languages or regions (Indic, Arabic, Southeast Asian, European, East Asian).32
Multimodal / vision-language modelsMultimodal / vision-language models covers models that take images/video/audio plus text (VLMs and omni models).30
Music & audio generationMusic & audio generation covers models and services that generate music and sound.24
Open-weight model familiesOpen-weight model families covers model families whose files can be downloaded and run on your own hardware.35
Realtime speech-to-speech modelsRealtime speech-to-speech models covers native audio-in/audio-out models and streaming APIs that converse in one real-time session (barge-in, low latency), replacing an STT-LLM-TTS cascade.18
Reasoning modelsReasoning models covers models trained with RL to think step-by-step before answering (test-time compute).26
RerankersRerankers covers models and APIs that re-order search results so the most relevant come first.20
Small / on-device LLMsSmall / on-device LLMs covers models built to run on phones, laptops, browsers and edge devices (roughly under 4B params).24
Speech (STT/TTS)Speech (STT/TTS) covers speech recognition, which turns speech into text, and speech synthesis, which turns text into speech.40
Time-series foundation modelsTime-series foundation models covers pretrained models for zero-shot forecasting and tabular prediction.16
World modelsWorld models covers interactive, action-conditioned video/3D world simulators for games, robotics and AV.17
Questions
Models, answered.
What is the Models layer of AI infrastructure?Models are the trained systems that do the work: language models, and models for speech, images, video, code, translation and forecasting. This layer covers both closed models reached through an API and open models you can download.
What does the Models layer include?It has 24 categories: 3D generation & motion, Avatar / digital humans, Background removal & matting, Benchmarks & leaderboards, Code models, Embedding models, Frontier model APIs, Image generation & editing, Image upscaling & restoration, Lip-sync & dubbing, Machine translation APIs, Model hubs / registries, Multilingual / regional-language models, Multimodal / vision-language models, Music & audio generation, Open-weight model families, Realtime speech-to-speech models, Reasoning models, Rerankers, Small / on-device LLMs, Speech (STT/TTS), Time-series foundation models, Video generation and World models.
How many Models products does Unfragile track?649 products across 24 categories, each category with its own page.
What works here
The three products on this layer.
SearchLive
RecommendNot live yet. This is the layer Recommend starts with.
RouteNot live yet.