Closed
Use the model through a service. The weights stay private.
Enterprise AI research
Models.
Explore AI models, the companies behind them, and the research to help you choose.
01 / The model landscape
Explore AI models.
See who makes each model, how you can access it, and what its license allows.
Open-weight
Download the weights and choose where to run them. Check the license.
Open-source
Access weights, training code, and data information, with broader rights to use and modify.
Open weights are one part of an open-source release.Read the OSI definition ↗
Sources checked 24 Oct 2026
11 models & model families
Frontier closed models
3Selected flagship families. Choose an exact version before testing.
| Model & developer | Access | License | Details and comparison |
|---|---|---|---|
| ClaudeAnthropic | Closed weightsHosted service | Proprietary | |
Where you can use itClaude API; selected versions through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Before you chooseCheck the exact version, platform availability, and data terms. Terms: Proprietary service terms Primary sourcesModel documentation | |||
| GeminiGoogle DeepMind | Closed weightsHosted service | Proprietary | |
Where you can use itGemini API and Google Cloud Vertex AI. Availability varies by version and region. Before you chooseCompare model versions and the commercial terms of each Google access route. Terms: Proprietary service terms Primary sourcesModel documentation | |||
| GPTOpenAI | Closed weightsHosted service | Proprietary | |
Where you can use itOpenAI API; selected models through Microsoft Foundry and Amazon Bedrock. Google Cloud access is in private preview. Before you chooseConfirm the exact version, region, and service terms. Google Cloud access is limited to approved preview customers. Terms: Proprietary service terms | |||
Open-weight & open-source models
8Released weights you can download, with different licenses and levels of openness.
| Model & developer | Access | License | Details and comparison |
|---|---|---|---|
| DeepSeek V4.1 FlashDeepSeek | Open weightsWeights + hosted APIs | MIT | |
Where you can use itRun the released weights yourself or use the DeepSeek API. Each hosting service has its own data and service terms. Before you chooseConfirm the checkpoint, serving hardware, data policy, and whether the API endpoint uses the same release. Terms: MIT | |||
| Gemma 4Google DeepMind | Open weightsDownloadable weights | Apache 2.0 | |
Where you can use itYour own deployment; Google Cloud Model Garden deployment options. Before you chooseCheck model size, hardware needs, license, and who operates the endpoint. Terms: Apache 2.0 | |||
| gpt-ossOpenAI | Open weightsDownloadable weights | Apache 2.0 | |
Where you can use itYour own deployment or a hosting provider. Not served in the OpenAI API. Before you chooseChoose a size and serving provider, then test quality and total operating cost. Terms: Apache 2.0; review usage policy Primary sourcesModel documentation | |||
| Granite 4.2IBM | Open weightsDownloadable weights | Apache 2.0 | |
Where you can use itYour own deployment. Check hosted availability for the exact variant. Before you chooseMatch the model size, reasoning settings, and serving setup to the task. Terms: Apache 2.0 | |||
| Mistral Large 3Mistral AI | Open weightsWeights + hosted APIs | Apache 2.0 | |
Where you can use itMistral API or your own deployment of the released weights. Confirm the model identifier and hosting terms. Before you chooseCheck hardware, serving support, and the exact release. Large 4 has a separate release schedule. Terms: Apache 2.0 | |||
| Muse Spark OpenMeta | Open weightsWeights + hosted APIs | Muse Model License | |
Where you can use itYour own deployment; listed on Microsoft Foundry, Google Cloud Model Garden, and Amazon SageMaker; served by third-party API hosts. Before you chooseRead the license terms before shipping a product on it. Plan hardware for a very large mixture-of-experts model. Terms: Muse Model License; attribution and revenue terms | |||
| Olmo 3Ai2 | Open-source releaseDownloadable weights | Apache 2.0 | |
Where you can use itDeploy a selected variant yourself. Ai2 publishes the training materials and checkpoints for inspection and adaptation. Before you chooseChoose the base, instruction-tuned, or reasoning variant. Assign an owner for hosting, evaluation, and updates. Terms: Apache 2.0 | |||
| Qwen3.8-27BAlibaba Cloud | Open weightsWeights; API listed | Apache 2.0 | |
Where you can use itSelf-host the released 27B weights. Qwen Cloud lists a hosted endpoint; confirm availability and region before planning production. Before you chooseThis record covers the 27B release. The model card still describes hosted access as coming soon, while Qwen Cloud lists an endpoint; confirm access directly. Terms: Apache 2.0 | |||
No models match these filters.
Try another name, developer, or license.
How to read this directory
This is a curated starting set, not a performance ranking. “Frontier” describes leading-edge capability; it is separate from whether weights are public. Family names identify where to begin. Confirm the exact version, hosting region, and terms before making a decision.
Release watch: Mistral Large 4 is available as a hosted preview. Its launch announcement schedules weights for the end of October; this directory lists the released Large 3 weights. Read the announcement ↗
Open-source labels describe release materials, not model quality or independent certification. View independent model results ↗
02 / Developers & providers
Meet the companies.
Some build models. Some run them for you. Many do both.
Anthropic
Model developerClaude models and hosted services, with selected models available through cloud partners.
View profileCohere
Model developerCommand models and deployment options for enterprise applications.
View profileGoogle Cloud
Developer & providerGemini, Gemma, and access to models from other developers.
View profileIBM
Developer & providerGranite models and enterprise AI software and services.
View profile
Foundry provides access to Microsoft and partner models, with deployment options and terms that vary by model.
View profileOpenAI
Model developerGPT models through hosted services, plus downloadable gpt-oss weights.
View profileSelected Institute company profiles. Explore the model records above for release-specific documentation.
Our evaluation approach03 / Research & evaluation
Make the choice concrete.
Define the work, agree on a test, and compare the full cost of a useful result.
Practical toolFree to use
Your model
buying brief.
Six questions to turn a list of models into a clear evaluation plan.
Build your briefDecision framework · Draft
Evaluate a platform.
Turn business requirements into questions you can test.
Cost analysis
Understand the full cost.
Look beyond token prices to the cost of completed work.
Capabilities & limits
Put model capability in context.
Understand how models fit into agents and real workflows.
Build a model buying brief
Write down the same requirements for every candidate. Your answers stay in this tab and are included in your download.
04 / Essential guides
Start with the fundamentals.
Clear explanations of the concepts behind the choices. Open to everyone.
Models
What are AI models?
Understand what a model does, how versions differ, and where it fits in your software.
Read the guideRunning a modelWhat is inference?
Learn what happens when a model runs and what affects the cost of a result.
Read the guideAccess & controlWhat are open-weight models?
Understand what you can download and run, and which limits still depend on the license.
Read the guide