AI Models

The frontier model race keeps reshuffling which model is worth paying for. GPT, Claude, and Gemini remain the three most used general purpose models, each pulling ahead on different tasks depending on the month, coding, writing, or reasoning benchmarks move around release to release rather than settling. Grok and DeepSeek have pushed hard on price and speed rather than chasing every benchmark, DeepSeek in particular narrowed the gap with much cheaper inference. Llama, Qwen, and Mistral are the open weight options, worth it if you want to self host, fine tune, or keep everything off someone else's servers. Gemini Nano is the outlier here, small enough to run on device rather than a data center.

Context window and price per token matter more day to day than leaderboard position for most use cases. If you're building something that calls a model thousands of times, the cheaper option often wins even at a slight quality cost. If you need a large context window for long documents, check that number specifically rather than assuming. Compare them below.