Kimi K2 Thinking
Kimi K2 ThinkingRankingin thinking mode?
Alternatives?
CheaperMiniMax M3thinkingOpen+11%-38%$0.51No strict JSON+1
BetterGPT-5.6 Lunamedium+2%-47%$0.016
OpenQwen3.8 2.4T A95Bthinking+12%+252%$2.16Memory (4-bit): 616 GB → 1,468 GB+5CheaperQwen3.8 FlashthinkingOpen+10%-73%$0.37max_tokens range · QwenCloud+5
CheaperGLM 5.3 FlashmaxOpen+9%-72%$0.28Defaults to max · Self-hosted+4MiniMax M3thinkingOpen+11%-38%$0.51No strict JSON+1Qwen3.8 FlashthinkingOpen+10%-73%$0.37max_tokens range · QwenCloud+5GLM 5.3 FlashmaxOpen+9%-72%$0.28Defaults to max · Self-hosted+4GPT-5.6 Lunamax+9%-47%$0.18
Agnes 2.5 Pro Beta+8%-82%$0.10 / $0.30GPT-5.6 Lunamedium+2%-47%$0.016
MiMo v2.5 ProthinkingOpen+3%-36%$0.054
Hy3thinkingOpen+7%-73%$0.072TInkling SmallthinkingOpen+7%-25%$0.089No strict JSONGPT-5.6 Terramedium+4%+428%$0.18Prompts with >272K input tokensQwen3.8 2.4T A95BthinkingOpen+12%+252%$2.16Memory (4-bit): 616 GB → 1,468 GB+5MiniMax M3thinkingOpen+11%-38%$0.51No strict JSON+1
DeepSeek V4 Pro 0813maxOpen+11%+132%$0.67Memory (4-bit): 616 GB → 990 GB+5Qwen3.8 FlashthinkingOpen+10%-73%$0.37max_tokens range · QwenCloud+5GLM 5.3maxOpen+9%+152%$2.01Reasoning cannot be disabled+3Kimi K3maxOpen+12%—$2.00Memory (4-bit): 616 GB → 1,668 GB+4Kimi K2.6thinkingOpen+9%—$0.80Kimi K2.7 CodethinkingOpen+7%—$0.54Instant mode is not supported · Self-hosted / vLLM / SGLangKimi K2.5thinkingOpen+5%—$0.60 / $3Kimi K2 0905non-thinkingOpen-9%—$0.40 / $2Kimi K2 0711non-thinkingOpen-9%—$0.50 / $2Context: 256K → 128K+1Kimi K3.1 Pron/a—$2 / $10Kimi K3.1 Pro Highspeedn/a—$4 / $20All Moonshot AI models →Run it yourself
estimateBenchmarks
all categoriesFAQ
How much does Kimi K2 Thinking cost?›
$0.47 per 1M input tokens and $2 per 1M output tokens (cheapest host: DeepInfra).
Is Kimi K2 Thinking free to use?›
There is no free API; the cheapest paid price is $0.47 per 1M input tokens. The weights are open, so you can run it yourself.
How good is Kimi K2 Thinking, and what is it best at?›
#103 of 334 in Chat & Knowledge, #58 of 239 in Coding, #30 of 242 in Agents & Tool Use, #40 of 95 in Writing & Language. Strongest at: Tool calling, Instructions.
Is Kimi K2 Thinking better than Gemini 3.5 Flash Lite?›
On GPQA Diamond, Kimi K2 Thinking scores 83.8 vs 83.8 — about the same.
How can I use Kimi K2 Thinking?›
Through the API of AWS Bedrock, DeepInfra, Featherless, Google, Novita.
What is the context window of Kimi K2 Thinking?›
256K tokens, with up to 98K output tokens.
When was Kimi K2 Thinking released? Is there a newer version?›
Released 2025-11-06. It has been superseded by Kimi K2.5.
Is Kimi K2 Thinking open source?›
Yes, the weights are open (modified-mit licence).
Can I run Kimi K2 Thinking locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Kimi K2 Thinking read images?›
No, it accepts text only.
What is Kimi K2 Thinking?›
Kimi K2 Thinking is an open-weights language model from Moonshot AI, released Nov 2025.
What are the best alternatives to Kimi K2 Thinking?›
MiniMax M3 (cheaper), GPT-5.6 Luna (better), Qwen3.8 2.4T A95B (open weights).