Google's Gemini 3.8 Flash and Moonshot AI's Kimi K3 compared on specs, pricing, and capabilities — as configured in AI Crucible ensembles.
| Specification | Gemini 3.8 Flash | Kimi K3 |
|---|---|---|
| Provider | Moonshot AI | |
| API model ID | gemini-3.8-flash | kimi-k3 |
| Description | Google's most intelligent Flash model, engineered for long-horizon software engineering with 1M context. | Moonshot AI's 2.8T-parameter flagship for long-horizon coding and knowledge work, with always-on reasoning, vision, and 1M context. |
| Context window | 1M tokens | 1.048576M tokens |
| Max output tokens | 65,536 | 8,192 |
| Input cost ($/1M tokens) | $0.90 | $3.60 |
| Output cost ($/1M tokens) | $4.50 | $18.00 |
| Cache read cost ($/1M tokens) | $0.09 | $0.36 |
| Latency | low | high |
| Reasoning model | Yes | Yes |
| Vision (image input) | Yes | Yes |
| Tool use | Yes | Yes |
See all model comparisons, full model specifications, benchmark results, or model comparison articles.