Moonshot AI's Kimi K3.1 Pro and DeepSeek's DeepSeek-V4-Pro compared on specs, pricing, and capabilities — as configured in AI Crucible ensembles.
| Specification | Kimi K3.1 Pro | DeepSeek-V4-Pro |
|---|---|---|
| Provider | Moonshot AI | DeepSeek |
| API model ID | kimi-k3.1-pro | deepseek-v4-pro |
| Description | Moonshot AI's most capable flagship, a 3T-parameter model for coding, agents, and deep reasoning, with vision and 1M context. | DeepSeek's most capable V4 model with dual thinking/non-thinking modes and 1M context. |
| Context window | 1.048576M tokens | 1M tokens |
| Max output tokens | 8,192 | 8,192 |
| Input cost ($/1M tokens) | $2.40 | $1.58 |
| Output cost ($/1M tokens) | $12.00 | $4.75 |
| Cache read cost ($/1M tokens) | $0.24 | $0.05 |
| Latency | high | medium |
| Reasoning model | Yes | Yes |
| Vision (image input) | Yes | No |
| Tool use | Yes | Yes |
See all model comparisons, full model specifications, benchmark results, or model comparison articles.