A 24-billion parameter mixture-of-experts model from LiquidAI that activates only 2 billion parameters per token for efficient inference.
- Strengths
- Balances model capacity with computational efficiency through sparse activation, enabling faster inference than dense models of similar scale.
- Best for
- On-device deployment and latency-sensitive applications where reducing inference compute is important.
- Limitations
- May underperform dense models on reasoning tasks or domains where its training data is sparse, due to its smaller active parameter count.
Input / 1M
$0.03
Output / 1M
$0.12
Cached input / 1M
—
Context window
32K
Price history
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 11 Jun 2026 | $0.03 | $0.12 | — | Imported from OpenRouter | openrouter.ai |