Large Language Models 🤖

The GPT model family (GPT-4o, GPT-5 and successors) - the single most-used AI model line in the world, powering ChatGPT and a huge share of AI features built into other products.

OpenAI's dedicated reasoning models (the "o-series") - trained to think step-by-step before answering, trading speed for much stronger performance on math, logic, and complex coding tasks.

OpenAI's image-generation models (DALL-E and GPT-image successors) - text-to-image generation built directly into ChatGPT and available via the OpenAI API.

Anthropic's mid-tier, balanced model line - the default choice for most Claude API usage and Claude Code, tuned to be strong at coding and long-context reasoning at a reasonable cost.

Google's fast, low-cost Gemini tier - optimized for high-volume, latency-sensitive applications where full Gemini Pro reasoning power isn't needed.

Google's reasoning-focused Gemini models (Gemini "Thinking" variants) - designed to work through multi-step problems before responding, competing directly with OpenAI's o-series.

DeepSeek's reasoning model line (DeepSeek-R1 and successors) - an open-weight reasoning model that shocked the industry by matching top closed-source reasoning models at a fraction of the training cost.

DeepSeek's general-purpose chat models (DeepSeek-V3 and successors) - strong all-around performance, fully open-weight, and dramatically cheaper to run than most closed-source competitors.

Meta's open-weight model family - free to download and run yourself (via Ollama, vLLM, etc.), the model most responsible for kickstarting the modern open-weight LLM ecosystem.

Mistral AI's model family - a French AI lab known for efficient, strong open-weight and API models, popular in the EU for data-sovereignty-conscious deployments.

Alibaba's open-weight Qwen model family - strong multilingual (especially Chinese/English) performance, widely used both directly and as a base for fine-tuned community models.

Microsoft's small language model (SLM) family - punches well above its parameter count, designed to run efficiently on modest hardware including laptops and edge devices.

xAI's Grok model family - built into X (Twitter), known for real-time access to X's own data and a more unfiltered, casual conversational style than most competitors.

Perplexity's search-grounded model - every answer comes with live web citations, built specifically for "AI search engine" style question-answering rather than general chat.

Amazon's own foundation model family, available through Amazon Bedrock - a lower-profile but deeply AWS-integrated option for teams already committed to that cloud ecosystem.

Cohere's enterprise-focused model - built with a heavy emphasis on retrieval-augmented generation (RAG) and enterprise search use cases rather than general consumer chat.

Reka AI's multimodal model family (Reka Flash and others) - a smaller lab building efficient multimodal models, the least-adopted model line in this ranking so far.