DeepSeek vs Kimi
Two open models with different centres of gravity. DeepSeek is reasoning-first and extremely cheap; Kimi K3 is the bigger all-rounder. Context is no longer the separator, DeepSeek V4 carries a million tokens too. If you are picking one open model to self-host, K3 is the more general answer, but DeepSeek remains hard to beat on cost per solved reasoning problem.
| DeepSeek | Kimi | |
|---|---|---|
| Provider | DeepSeek | Moonshot AI |
| Current release | V4 | K3 |
| Tier | Reasoning | Open weights |
| Context | 1M | 1M |
| Cost | Low cost | Low cost |
| Modalities | text | text, vision |
| Open weights | Yes | Yes |
Cost is a tier, not a quote, providers change prices often. Check DeepSeek and Moonshot AI before you commit. Last updated 2026-09-04.
Which one to pick
DeepSeekDeepSeek · V4
Open reasoning at a fraction of the cost of closed reasoning models
Reach for it when- Chain-of-thought reasoning on a tight budget
- Smaller footprint to run
KimiMoonshot AI · K3
Frontier-adjacent quality you can self-host, the open model closest to the closed leaders
Reach for it when- One open model to cover most workloads
- Broad general capability rather than reasoning alone