deepseek-v4-pro vs kimi-k3
| deepseek-v4-pro | kimi-k3 | |
|---|---|---|
| Providers | DeepSeek | Moonshot AI |
| Input $/M | $0.66 | $1.54 |
| Output $/M | $1.98 | $7.69 |
| Output multiplier | ×3.00 | ×5.00 |
| Context window | 1M tokens | 1M tokens |
| Intelligence index | 36.0 | 43.6 |
| Coding index | 68.8 | 76.2 |
What the numbers mean
deepseek-v4-pro is roughly 2.3× cheaper on input than kimi-k3. If your workload passes the quality bar on the cheaper model, that gap compounds across every request; route to kimi-k3 deliberately rather than by default.
On measured intelligence index the gap is 7.6 points, with kimi-k3 ahead. That is a large enough spread to be visible on multi-step tasks.
Tags: deepseek-v4-pro — Chat, Vision, Reasoning, Coding, 工具调用, Long context, 超大输出 · kimi-k3 — Chat, Vision, Reasoning, Coding, 工具调用, Long context, 超大输出
Code
from openai import OpenAI
client = OpenAI(base_url="https://starseaapi.com/v1", api_key="sk-...")
# deepseek-v4-pro
client.chat.completions.create(model="deepseek-v4-pro", messages=[{"role":"user","content":"Hi"}])
# kimi-k3
client.chat.completions.create(model="kimi-k3", messages=[{"role": "user", "content": "Hi"}])