AI & models
The right model
for the work.
I build with OpenAI, Claude and open-weight models. I compare them against the actual task, including response quality, speed, cost and data requirements.
The current catalogue
Names below are checked automatically against provider catalogues. Availability for a project depends on the provider and the account.
OpenAI
Provider modelsProvider catalogueChecked
Claude
Anthropic modelsProvider catalogueChecked
Open weights
Self-hostable modelsProvider catalogueChecked
The catalogue is checked daily. If a source is unavailable, its last verified list stays visible with the original check date. It records available models; each project has its own agreed model configuration.
Microsoft Foundry
For work in Azure, Microsoft Foundry brings model access, agents and evaluation together. It supports the identity and access controls a team may already use through Microsoft Entra.
I consider Foundry when managed deployment, evaluation and operational controls need to sit within the organisation’s Azure setup. Direct provider APIs can suit smaller, focused applications.
Microsoft’s Foundry documentationOpen weights.
More control.
Open-weight models can run on infrastructure chosen for the project. That can help with deployment control and data handling, but the hosting, hardware and model licence still need consideration.
I evaluate the model on representative examples before deciding it belongs in a product. A release announcement is the beginning of that evaluation.
Discuss an AI project