DeepSeek V4 vs Kimi K2 Thinking

Benchmarks, API pricing and specs, head to head. Data updated 2026-06-10.

DeepSeek V4

DeepSeek · Mar 2026

The open-weights shock of 2026: ~81% SWE-bench Verified, a 1M context and MIT license at $0.30/$0.50 per million tokens.

Kimi K2 Thinking

Moonshot AI · Nov 2025

67.6

A trillion-parameter open reasoning agent that can chain 200–300 tool calls — the open-weights agentic standout of late 2025.

The verdict

DeepSeek V4 wins 1 of the 1 benchmarks these models share, against 0 for Kimi K2 Thinking. On price the gap is dramatic: DeepSeek V4 works out roughly 3.1x cheaper per blended million tokens. DeepSeek V4 also takes 1M of context versus 262K for Kimi K2 Thinking.

Benchmark head-to-head 10

SWE-bench Verified
~81%71.3%
DeepSeek V4Kimi K2 Thinking

Specs & pricing

DeepSeek V4Kimi K2 Thinking
modhub Index67.6
Input price / 1M$0.3$0.6
Output price / 1M$0.5$2.5
Context window1M262K
Max output64K
Open weightsyes (MIT)yes (Modified MIT)
Reasoning modelyesyes
Multimodal inputtexttext
Knowledge cutoffDec 2025Apr 2025
ReleasedMar 2026Nov 2025
Example monthly cost*$3.75$9.75

* 10M input + 1.5M output tokens per month at list prices, no caching. Green = better value on that row.

Frequently asked questions

Which is better, DeepSeek V4 or Kimi K2 Thinking?
DeepSeek V4 wins 1 of the 1 benchmarks these models share, against 0 for Kimi K2 Thinking. On price the gap is dramatic: DeepSeek V4 works out roughly 3.1x cheaper per blended million tokens. DeepSeek V4 also takes 1M of context versus 262K for Kimi K2 Thinking.
Which is cheaper, DeepSeek V4 or Kimi K2 Thinking?
DeepSeek V4 costs $0.3/$0.5 per million input/output tokens, while Kimi K2 Thinking costs $0.6/$2.5. For a typical workload of 10M input and 1.5M output tokens per month, that's $3.75 versus $9.75.
Which model is better for coding, DeepSeek V4 or Kimi K2 Thinking?
On SWE-bench Verified — the standard agentic-coding benchmark — DeepSeek V4 scores ~81% versus 71.3% for Kimi K2 Thinking, making DeepSeek V4 the stronger pick for coding agents.

More comparisons