Probed: partially recovered the spoken codeword. Treat sound commentary as unverified.
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 | — |
| completion | 0 | — |
{
"temperature": 1,
"top_p": 0.95,
"top_k": 64
}