Probed: missed a spoken codeword inside the clip. Any comment it makes on sound is invented.
SHAPE: square, yellow AUDIO: NONE
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...
| Field | Per token (USD) | Per million |
|---|---|---|
| prompt | 0.0000000765 | $0.076 |
| completion | 0.000000255 | $0.255 |
| input_cache_read | 0.0000000425 | $0.042 |
{
"temperature": 1,
"top_p": 0.95,
"top_k": 64
}10 answersin this database. A round opens every model's answer to the same prompt.