TL;DRGRUs are a simpler, faster gated RNN — fewer gates than an LSTM, often comparable results.
The Gated Recurrent Unit merges the LSTM's gates into two (reset and update) and drops the separate cell state. Fewer parameters means faster training and less data, often matching LSTM performance on many tasks. When you need a gated RNN, GRU is the lighter default; LSTM the heavier option for the longest dependencies. Both were largely superseded by transformers.
Key points
Two gates (reset, update), no separate cell state
Fewer parameters → faster, less data
Often matches LSTM performance
Lighter default gated RNN
Common mistakes
Assuming GRU is always worse than LSTM
Using either where a transformer fits
Over-tuning RNN choice on small gains
Try it: Contrast a GRU and an LSTM in gate count and complexity.
Example code
<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>LSTM: 3 gates + cell state (heavier)
GRU: 2 gates, no cell state (lighter, faster)
often comparable accuracy</pre></body></html>