Next Token Visualizer
Explore how language models predict the next word
How it works
Language models predict text one piece at a time. Given your input, the model calculates the probability of every possible next word. The selected word is what the model chose based on these probabilities. Click any prediction to add it and keep going!
?
Temperature reshapes the probability distribution before sampling. At 0, the highest-probability token is always selected. Higher values flatten the distribution, giving lower-probability tokens more chance.
?
Top P (nucleus sampling) limits the candidate pool before sampling. At 0.9, only tokens within the top 90% cumulative probability are considered. This prevents sampling very unlikely tokens even at high temperature.
Powered by OpenRouter API