Best AI models in 2026
Language models you can call from an app, from hosted flagships to open weights. 21 tools, ranked by live GitHub and npm numbers.
| Tool | Pricing | GitHub stars | npm a week | Last release |
|---|---|---|---|---|
| DeepSeek V3Open-weight large language model from DeepSeek, since superseded by the DeepSeek V4 family. | Open source | 105k | None | 1 year ago |
| DeepSeek R1Open-weight reasoning model from DeepSeek, since superseded by the DeepSeek V4 family. | Open source | 91.9k | None | 1 year ago |
| Llama 2Meta's 2023 open-weight language model family, superseded by newer Llama releases. | Open source | 59.6k | None | None |
| StarCoderOpen source code-generation model from the BigCode project, trained on 80+ programming languages. | Open source | 7.5k | None | None |
| Gemma 4Google DeepMind's current generation of lightweight, open-weight language models. | Open source | 5.8k | None | 4 months ago |
| DeepSeek V4DeepSeek's current open-weight model family, including V4.1-Flash and V4-Pro. | Open source | None | None | None |
| DevstralMistral AI's open-weight model built specifically for coding and software engineering agents. | Open source | None | None | None |
AI Models comparisons
Alternatives
How to choose
This category lists individual models and model families: the thing you name in an API call or download and run. Hosted models are reached through the maker's API or a cloud platform and are billed per token. Open-weight models can be downloaded and run on your own hardware or through a hosting provider.
Models are replaced often. Each page here says whether a model is current, superseded or retired, and points to its successor when there is one, so check the status before you build on a model.
When you compare models, look at the price per million input and output tokens, the context window, whether the weights are open, and which platforms serve it.
What to compare
- Status: current, superseded or retired
- Price per million input and output tokens
- Context window
- Open weights or API only, and the license
- Where it is available: maker API, cloud platforms, local runners
- What it is built for: general use, coding, reasoning
Questions
What does open-weight mean?
The trained model files are published, so you can download and run the model yourself. It is not the same as open source: each model has its own license, and some restrict commercial use.
What is a context window?
It is the amount of text, measured in tokens, that a model can take in at once, including your prompt, any documents and the conversation so far. A larger window lets the model work across longer documents or codebases in one request.
Why are some models marked superseded or retired?
Superseded means the maker has released a newer model that replaces it, though the old one may still be served. Retired means it has been switched off. Pages for both are kept so that older references still resolve, and each links to the current model.