DeepSeek V4 Pro 0813 vs Kimi K3
DeepSeek V4 Pro 0813 vs Kimi K3: which AI model is cheaper, has the larger context window and more capabilities? Side-by-side comparison updated daily.
Quick verdict
- DeepSeek V4 Pro 0813 is 6.1× cheaper than Kimi K3 for the same mix of input and output tokens.
- Both have the same context window.
- DeepSeek V4 Pro 0813 is the more recent release (Aug 12, 2026).
- Only Kimi K3 accepts images as input.
- Kimi K3 scores higher on independent tests (overall index 157.7 vs 155.4).
- Choose DeepSeek V4 Pro 0813 if cost matters most; choose Kimi K3 if you need its specific strengths above. For the best results, test both on your own prompts.
| DeepSeek V4 Pro 0813 | Kimi K3 | |
|---|---|---|
| Company | DeepSeek | Moonshot AI |
| Released | Aug 12, 2026 ● | Jul 16, 2026 |
| Overall index | 155.4 | 157.7 ● |
| SWE-bench Verified | 77.6% | — |
| GPQA Diamond | 91.7% | 93.1% ● |
| AIME math | 98.6% ● | 97.2% |
| FrontierMath | 64.6% | 72.2% ● |
| SimpleQA Verified | 52.9% ● | 50.6% |
| Input (per 1M tokens) | $0.66 ● | $3 |
| Output (per 1M tokens) | $1.98 ● | $15 |
| Cost for 1M input + 1M output tokens | $2.64 ● | $18.00 |
| Providers | 20 | 22 ● |
| Context | 1M | 1M |
| Max output | 384K | 944K ● |
| Inputs | Text | Text, Image, Video ● |
| Outputs | Text | Text |
| Reasoning | ✓ | ✓ |
| Tool use | ✓ | ✓ |
| Structured output | ✓ | ✓ |
| Open weights | ✓ | ✓ |