DeepSeek V4 Pro 0813 versus Kimi K3

Two models head to head

This comparison pits two open-weight models that you can install on your own servers against each other: DeepSeek V4 Pro 0813 and Kimi K3, one from DeepSeek, the other from Moonshot AI. The first dates from August 13, 2026, the second from July 16, 2026.

All figures below come from public sources checked on October 22, 2026: models.dev and OpenRouter for prices and capabilities, Epoch AI for scores.

What the benchmarks say

On our capability index, DeepSeek V4 Pro 0813 scores 72.2 out of 100 and Kimi K3 80.7. Kimi K3 leads by 8.5 points.

On the 5 benchmarks reported for both, DeepSeek V4 Pro 0813 wins 1 time and Kimi K3 4:

• GPQA Diamond (PhD-level science questions): DeepSeek V4 Pro 0813 91.7%, Kimi K3 93.1%
• OTIS Mock AIME (competition mathematics problems): DeepSeek V4 Pro 0813 98.6%, Kimi K3 97.2%
• FrontierMath (very hard research-level mathematics): DeepSeek V4 Pro 0813 64.6%, Kimi K3 72.2%
• FrontierCode (advanced programming tasks): DeepSeek V4 Pro 0813 28.6%, Kimi K3 44.2%
• APEX-Agents (professional tasks carried out by an agent): DeepSeek V4 Pro 0813 47.3%, Kimi K3 47.8%

Which one costs less?

DeepSeek V4 Pro 0813 costs $0.66 per million input tokens and $1.98 per million output tokens; Kimi K3 costs $3 and $15.

For a typical agent task (read 20,000 tokens, write 2,000), expect $0.017 with DeepSeek V4 Pro 0813 versus $0.09 with Kimi K3. DeepSeek V4 Pro 0813 is 5.2 times cheaper, a gap that adds up quickly over 1,000 tasks: $72.84 difference.

Run the numbers with your own volumes in the agent cost calculator.

Context and capabilities

Both models have the same context window: 1,048,576 tokens, roughly 1,573 pages.

Reasoning mode: DeepSeek V4 Pro 0813 yes, Kimi K3 yes. Tool calling: DeepSeek V4 Pro 0813 yes, Kimi K3 yes. Image input: DeepSeek V4 Pro 0813 no, Kimi K3 yes.

Both publish their weights: you can host them yourself.

In short

Our reading of the data:

• Choose DeepSeek V4 Pro 0813 for a lower cost.
• Choose Kimi K3 for a higher capability index and image input.

This verdict relies on public figures; it does not replace a trial on your own tasks.