A model generates text, images or other outputs from your input. Start with where you want to run it and what you can spend.
These groups overlap. Open-weight describes access to the files; local or cloud describes where they run. Mistral Large 4 has both a hosted API and server weights; Qwen3.8-27B spans servers and high-memory personal machines. “Frontier” is an editorial capability label, not a hosting type or a measured ranking.
Use the provider’s computers through an application or a developer API. No model-sized GPU is required on your device. Compare account access and billing: a chat subscription and API usage are separate routes.
Includes editorial frontier candidates and efficient options such as Haiku. The prices below are base USD API rates checked 29 October 2026, with key conditions alongside them; they are not chat-plan prices.
Back to all three groupsCloud API model gpt-6.1-sol for coding, document understanding and multi-step professional work.
License Proprietary hosted service
Cloud API model gpt-6.1-terra for visual understanding, computer use and writing; released October 26.
License Proprietary hosted service
A commercial cloud model for complex agentic coding and knowledge work, announced September 22.
License Proprietary hosted service
General coding and knowledge work through the Claude API, with documented text/image input and text output.
License Proprietary hosted service
High-volume classification and extraction through the Claude API, with documented text/image input and text output.
License Proprietary hosted service
Demanding reasoning and long-horizon work through the Claude API, with documented text/image input and text output.
License Proprietary hosted service
A cloud model accepting text, images, video, audio and PDFs and producing text; computer use remains a preview.
License Proprietary hosted service
A large text-and-image model available through Mistral's paid API or as downloadable weights for organizations operating substantial GPU infrastructure.
License Custom Mistral open-weight license; commercial conditions
These downloadable models typically call for substantial GPU capacity, often rented in the cloud. Open-weight describes access to trained files, not unrestricted licensing or free hosting. High-memory machines you own can also be an option.
Back to all three groupsA large text-and-image model available through Mistral's paid API or as downloadable weights for organizations operating substantial GPU infrastructure.
License Custom Mistral open-weight license; commercial conditions
A downloadable text model for self-managed reasoning and tool-use applications, with a substantially larger memory footprint than gpt-oss-20b.
License Apache-2.0
A downloadable vision-language model for text, image and video understanding with configurable thinking for agentic workflows.
License Apache-2.0
A downloadable text-to-image model designed to generate images in a small number of inference steps.
License Apache-2.0; download access conditions
These are smaller, compressed or specialist models to consider for your own machine. Read the format and memory figures first: download size is not RAM or VRAM, and longer conversations need additional memory. The 27B option needs more headroom than the compact baselines.
Back to all three groupsA downloadable text model with switchable thinking and non-thinking modes for local chat, reasoning and tool-using experiments.
License Apache-2.0
An instruction-tuned, open-weight model that accepts text and images and produces text for compact multimodal applications.
License Gemma Terms; usage-license acceptance required
OpenAI's downloadable mixture-of-experts text model supports configurable reasoning effort and tool-oriented application development.
License Apache-2.0
A smaller reasoning-focused model distilled from DeepSeek-R1 and based on the Qwen 2.5 family.
License MIT release; Qwen-derived base
A downloadable speech-recognition model optimized from Whisper large-v3 for faster transcription.
License MIT
A downloadable instruction-tuned multimodal model accepts text, image and audio inputs and returns text for on-device-oriented applications.
License Apache-2.0
A downloadable vision-language model for text, image and video understanding with configurable thinking for agentic workflows.
License Apache-2.0
The source snapshots below contain public model repositories hosted on Hugging Face. A model repository can contain a model, a variant or a community conversion; it is not necessarily a distinct commercial product. These lists do not measure use of hosted Claude, GPT or Gemini services.
Trending preserves the provider’s trending-score order. Most downloaded uses trailing-30-day query-file request counts, not unique users; embeddings and retrieval models may lead that view. Use the task filter to find the kind of input and output you need, then read the publisher’s model card. The grouped model profiles above explain selected cloud and downloadable models in more depth.
All 50 returned public model repositories across returned tasks, in official API order.
Produce text from image and text inputs, such as descriptions or answers about an image.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Generate or continue text from a prompt. Check the card for the intended languages, context and deployment.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Generate or continue text from a prompt. Check the card for the intended languages, context and deployment.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Generate or continue text from a prompt. Check the card for the intended languages, context and deployment.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Generate or continue text from a prompt. Check the card for the intended languages, context and deployment.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Produce text from image and text inputs, such as descriptions or answers about an image.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Generate or continue text from a prompt. Check the card for the intended languages, context and deployment.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Create motion from an image input. Check supported formats and inference requirements.
Open the publisher’s model cardSource discovery entry; not independently reviewed or endorsed.Provider-defined trendingScore; not growth or quality.
Source order is retained when you filter. This is a dated capture, not continuous tracking. Community descriptions and cards are not independently validated product recommendations. Read the full method.
