The open-model ecosystem is rich and develops fast. How to navigate and choose: 1) Model diversity: — By task type: language models (LLMs for text), embedders (for semantic search/RAG), image models (generation, CV), speech (recognition/synthesis), multimodal. For your task there are almost certainly open models. — By size: from compact (fast, modest requirements, can run even on an ordinary computer/laptop) to large (more powerful but requiring serious hardware, usually a GPU). Size is a key selection parameter (more on this later). — By specialization: universal models and ones tuned for the specific (code, a certain language, a domain). 2) Where to get them: open models are published in model hubs — community platforms where they can be found, the description viewed and downloaded. There too is model info (size, purpose, requirements, license), often examples and discussions. These are like "stores" of open models. 3) How to choose a model (practical criteria): — Task: what the model is for (an LLM for text? an embedder? for images?). Take a model for your task type. — Size vs your resources: KEY. A large model is more powerful but requires more memory/computation (often a GPU). Make sure the model fits your hardware. Often it is sensible to take a smaller model that copes with the task and runs on available hardware (not chasing the biggest). — Quality for the task: not "the most powerful in general" but "good enough for YOUR task". For many tasks a compact open model suffices. — Language: make sure the model works well with your language — some open models are stronger in English (as discussed for embedders in the vector-search course). — License (IMPORTANT): open models have licenses defining how they can be used (including commercially). Not all "open" models can be freely used in business — CHECK the license before commercial use. This is an important legal point (a link to the law course). — Reputation/support: popular, well-documented models with an active community are easier to use and more reliable. 4) The gap with closed ones is narrowing: previously open models were noticeably inferior to closed, now the gap is decreasing, and good open models are quite competitive for many practical tasks. Do not consider open = "worse" — for your task an open one may be an excellent choice (and with the privacy/control/cost pluses from the previous lesson). 5) Ready tools for running: the ecosystem includes not only models but tools easing their download and running (the next module). You do not need to be a researcher — there are convenient ways to take a ready open model and use it. Practical rule: the open-model ecosystem is rich — for most tasks there are quality open models of various sizes, available in hubs. Choose for the TASK and for YOUR RESOURCES (model size vs your hardware — the key balance), check LANGUAGE support and necessarily the LICENSE (can it be used commercially). Often a compact open model that copes with the task on available hardware is a better practical choice than chasing the biggest. The gap with closed models is narrowing, so open ones are a real, not a "backup" option.