AI Models
Use these notes to narrow a technology choice, not to find a permanent winner. Prices, access, and hardware support can change after a note’s snapshot date; check that date and the tested configuration before relying on a comparison.
Choose a hosted model or API
Model & API Radar compares hosted model tiers, API access, free inference options, and prices. Start here if you want to use a remote service rather than run a model on your own hardware.
Run a model locally
- Will it fit? Single-GPU local models examines model artifacts, quantization, and memory headroom. Fitting in memory alone does not establish useful speed, context length, or task quality.
- Which coding model should I try? Local coding models narrows candidates for repository tasks, with attention to the exact model artifact and prompt template.
- Which engine should run it? Local inference runtimes helps match the serving software to hardware and workload.
- Does the whole setup work for my tasks? Local model evaluation covers repeatable tasks, recorded configurations, failures, and uncertainty.
Choose a coding agent
Free allowances for AI coding CLIs distinguishes login grants, subscription usage, API billing, and promotions. Coding agent evaluation compares products and explains how to try them on repository work. For the underlying design rather than product choice, read agent loops for stopping and control, or tool contracts for safe tool interaction.
Read comparisons in context
A provider’s model-card numbers are not independent measurements, and a leaderboard result belongs to a particular model, test harness, version, and configuration. An omitted candidate is unknown, not necessarily worse. Evidence and bias in AI notes explains how to judge these differences.
Coverage is strongest for terminal coding, English documentation, and NVIDIA or consumer local hardware. It is thinner for browser agents, multimodal creation, non-English evaluation, enterprise governance, energy use, and accessibility; the comparisons may not answer those needs.