Kimi K2 Thinking vs Llama 4 Maverick

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

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.

Llama 4 Maverick

Meta · Apr 2025

Meta's natively multimodal 400B MoE — the largest openly downloadable US-made model, served cheaply across many hosts.

The verdict

Kimi K2 Thinking wins 2 of the 2 benchmarks these models share, against 0 for Llama 4 Maverick. Llama 4 Maverick is about 2.6x cheaper per blended million tokens (3:1 input:output mix). Llama 4 Maverick also takes 1M of context versus 262K for Kimi K2 Thinking.

Benchmark head-to-head 20

GPQA Diamond
84.5%69.8%
Kimi K2 ThinkingLlama 4 Maverick
MMLU-Pro
~84.6%80.5%
Kimi K2 ThinkingLlama 4 Maverick

Specs & pricing

Kimi K2 ThinkingLlama 4 Maverick
modhub Index67.6
Input price / 1M$0.6$0.27
Output price / 1M$2.5$0.85
Context window262K1M
Max output
Open weightsyes (Modified MIT)yes (Llama 4 Community License)
Reasoning modelyesno
Multimodal inputtexttext, image
Knowledge cutoffApr 2025Aug 2024
ReleasedNov 2025Apr 2025
Example monthly cost*$9.75$3.98

* 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, Kimi K2 Thinking or Llama 4 Maverick?
Kimi K2 Thinking wins 2 of the 2 benchmarks these models share, against 0 for Llama 4 Maverick. Llama 4 Maverick is about 2.6x cheaper per blended million tokens (3:1 input:output mix). Llama 4 Maverick also takes 1M of context versus 262K for Kimi K2 Thinking.
Which is cheaper, Kimi K2 Thinking or Llama 4 Maverick?
Kimi K2 Thinking costs $0.6/$2.5 per million input/output tokens, while Llama 4 Maverick costs $0.27/$0.85. For a typical workload of 10M input and 1.5M output tokens per month, that's $9.75 versus $3.98.
Which model is better for coding, Kimi K2 Thinking or Llama 4 Maverick?
We don't yet track SWE-bench Verified results for both models; check their individual pages for coding-related scores.

More comparisons