A 120B-parameter open hybrid MoE model from NVIDIA that activates 12B parameters per token, combining Mamba and Transformer architectures for inference efficiency.
- Strengths
- Delivers strong reasoning and accuracy on complex tasks while maintaining low per-token compute costs through mixture-of-experts sparsity.
- Best for
- Multi-agent systems and complex reasoning tasks where you need model capability without proportional compute overhead.
- Limitations
- As an open model requiring self-hosting, it demands significant infrastructure; effectiveness on specialized domains depends on fine-tuning.
Input / 1M
$0.085
Output / 1M
$0.4
Cached input / 1M
—
Context window
262K
Price history
Input (solid)Output (dashed)
Price change
- 30d
- in decreased 0.0% out decreased 0.0%
- 90d
- in decreased 5.6% out decreased 11.1%
- 1y
- in decreased 5.6% out decreased 11.1%
- Since launch
- in decreased 5.6% out decreased 11.1%
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 25 Jul 2026 | $0.085 | $0.4 | — | Imported from OpenRouter | openrouter.ai |
| 21 Jul 2026 | $0.08 | $0.45 | — | Imported from OpenRouter | openrouter.ai |
| 19 Jul 2026 | $0.085 | $0.4 | — | Imported from OpenRouter | openrouter.ai |
| 15 Jul 2026 | $0.21 | $0.455 | $0.06 | Imported from OpenRouter | openrouter.ai |
| 4 Jul 2026 | $0.08 | $0.45 | — | Imported from OpenRouter | openrouter.ai |
| 26 Jun 2026 | $0.085 | $0.4 | — | Imported from OpenRouter | openrouter.ai |
| 11 Jun 2026 | $0.09 | $0.45 | — | Imported from OpenRouter | openrouter.ai |