DeepSeek V4-Pro vs Kimi K3
DeepSeek V4-Pro is from DeepSeek, released August 13, 2026. Kimi K3 is from Moonshot AI, released July 16, 2026.
- LMArena Elo1458 vs 1487. Kimi K3 is 2.0% ahead on LMArena Elo.
- Input per 1M$1.32 vs $3. DeepSeek V4-Pro costs 56% less per million input tokens.
- Output per 1M$3.96 vs $15. DeepSeek V4-Pro costs 74% less per million output tokens.
- Context window1M vs 1M. Both hold 1M tokens of context.
Prices are the vendors’ published API rates; Elo is LMArena’s text board. Each figure is dated on its model page.
Interactive Comparison
Loading chart...
$1.32 / $3.96 per million tokens | $3 / $15 per million tokens | |
|---|---|---|
| Identity | ||
| Developer | DeepSeek | Moonshot AI |
| Released | Aug 2026 | Jul 2026 |
| Status | GA | GA |
| Licence | MIT | Open weights (Kimi K3 License) |
| Self-hostable | Yes - open weights on Hugging Face | Yes |
| Cost | ||
| Blended $/1M tokensinput × 0.75 + output × 0.25 | $1.980 / 1M tokens | $6.000 / 1M tokens |
| Input price | $1.320 / 1M tokens | $3.000 / 1M tokens |
| Output price | $3.960 / 1M tokens | $15.000 / 1M tokens |
| Cached input | $0.044 / 1M tokens (peak); $0.022 off-peak | $0.30 / 1M tokens (automatic context-cache hit; $3.00 on a cache miss) |
| Batch discount | Off-peak billing: 50% off all rates outside 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday (weekends and Chinese public holidays are off-peak in full) | Not offered |
| Free tier | Not offered | Not offered |
| Capacity | ||
| Context window | 1M tokens | 1M tokens |
| Max output | 384k tokens | 128k tokens (default; max_completion_tokens configurable up to 1M) |
| Long-context surcharge | None (flat rate across the 1M window) | None (flat rate across the 1M window) |
| Capability | ||
| Vision in | No | Yes |
| Audio in | No | No |
| Function calling | Yes | Yes |
| Structured output | Yes | Yes |
| Extended reasoning | Yes - thinking effort levels low, high and max | Yes |
| Web search | No | Yes (tool currently being updated by Moonshot; not recommended for near-term use) |
| Code execution | No | Not offered |
| Access | ||
| API | Yes | Yes |
| Chat app | Yes | Yes |
| Cloud marketplaces | Not offered | Azure AI Foundry (via Fireworks AI), OpenRouter |
| Fine-tuning | Not offered | Not offered |
| Measured quality | ||
| LMArena Elo | 1458 (checked Oct 2026) | 1487 (checked Oct 2026) |
| LMArena rank | Rank 64 | Rank 18 |
| GPQA Diamond | 91 | 93 |
| SWE-bench Verified | 78 · max effort | — |
| Elo per dollarLMArena Elo ÷ blended $/1M tokens | 736 | 248 |
