TOOLS · LLM GUIDE
Large language models, explained for builders
Which model families exist, who makes them, whether you can download them, and where to try them free. Then test them yourself — that is the only benchmark that matters for your work.
| Family | Maker | Weights | Try free | Good at | Note |
|---|---|---|---|---|---|
| Llama | Meta | Open weights (licence terms apply) | Groq, OpenRouter, Ollama | General chat, summarising, coding help | Many sizes; small ones run locally. |
| Gemini | Closed (Gemma is the open family) | Gemini app, Google AI Studio free key | Long documents, multimodal input, multilingual | Flash models are the fast, cheap tier. | |
| Gemma | Open weights | Ollama, LM Studio, some APIs | Local and on-device use | Small enough for laptops in smaller sizes. | |
| GPT | OpenAI | Closed (gpt-oss models are open-weight) | ChatGPT free plan; gpt-oss on Groq/OpenRouter | General assistant tasks, tool use | gpt-oss models can run on other providers. |
| Claude | Anthropic | Closed | Claude free plan | Careful long-form writing and analysis | API is paid. |
| Mistral | Mistral AI | Mix of open and closed | Mistral free Experiment plan, Le Chat | Efficient models, European languages, coding | Phone verification for the free API plan. |
| Qwen | Alibaba | Open weights (most sizes) | Groq, OpenRouter, Ollama | Coding, maths, multilingual including Chinese | Strong small models. |
| DeepSeek | DeepSeek | Open weights | OpenRouter, Ollama | Reasoning, maths and coding | Check data terms of hosted versions. |
| Phi | Microsoft | Open weights | Ollama, LM Studio, Azure | Small, efficient reasoning | Designed for low-resource devices. |
| Command | Cohere | Mix (some open for research) | Cohere trial key | Retrieval (RAG) and business text | Trial keys are for testing, not production. |
| Grok | xAI | Mostly closed | Inside X (limits apply) | Chat with access to X posts | Availability depends on account. |
| Kimi | Moonshot AI | Mix of open and closed | OpenRouter, Groq (some versions) | Long context, agentic coding | Check which version a provider hosts. |
How to choose a model in 20 minutes
- Write three prompts from your real work.
- Run each on two or three models (free keys from our guides).
- Score each answer: correct, complete, clear, follows format (0–2 each).
- Note speed and any errors or refusals.
- Pick the cheapest model that scores well — not the most famous.