AI model matchup

Kimi K2-Thinking-0905 vs DeepSeek-V4-Flash-0423

A side-by-side look at published capability scores, listed API pricing, speed, context, and license - to help you decide which trade-offs fit your work.

OpenAI

GPT-5.6 Sol

Rank #1 • Score 57.2
Comparison Metrics
Speed: 89c/s vs 68c/s
Pricing: $7.78 vs $7.22
Anthropic

Claude Opus 5

Rank #2 • Score 56.2

Kimi K2-Thinking-0905vsDeepSeek-V4-Flash-0423

DeepSeek-V4-Flash-0423 leads the RankLLMs Index by 0.2 points (36.6 vs 36.4). Verified API Pricing: Free/Open Source (Kimi K2-Thinking-0905) vs Free/Open Source (DeepSeek-V4-Flash-0423) per 1M tokens.

Capability profile

Kimi K2-Thinking-0905DeepSeek-V4-Flash-0423

Speed duel

tps
Kimi K2-Thinking-090550
DeepSeek-V4-Flash-042364

Price duel

$/1M
Kimi K2-Thinking-0905Free/Open Source
DeepSeek-V4-Flash-0423Free/Open Source

Shorter bar = cheaper. Blended per 1M tokens.

MetricKimi K2-Thinking-0905DeepSeek-V4-Flash-0423
RankLLMs Overall Score36.4 pts 36.6 pts ▲
Reasoning36.6 37.6 ▲
Coding21.7 27.9 ▲
Agentic Tool Use16.6 17.3 ▲
Code Arena Elo941 null
Inference Speed50 tps 64 tps ▲
API Price (per 1M tokens)Free/Open Source Free/Open Source
Context Window128K 1M
LicenseOpen Source Open Source
Global Rank#78 #77 ▲

Decision notes

What the numbers suggest

DeepSeek-V4-Flash-0423 leads the RankLLMs overall score by 0.2 points. Kimi K2-Thinking-0905 has the lower listed API rate (Free/Open Source per 1M tokens).

Treat the overall index as a starting point, then prioritize the benchmark that resembles your task. Published scores and provider pricing can change; this comparison does not represent a first-party test of either model.

Overall index
DeepSeek-V4-Flash-0423 leads
36.4 to 36.6 pts
Coding score
DeepSeek-V4-Flash-0423 leads
21.7 to 27.9 pts
Reasoning score
DeepSeek-V4-Flash-0423 leads
36.6 to 37.6 pts
Agentic score
DeepSeek-V4-Flash-0423 leads
16.6 to 17.3 pts

Choose by workload

Which profile fits your use?

Moonshot AI

Consider Kimi K2-Thinking-0905 when

your workload gives more weight to the lower listed API rate. Compare those signals with your own task mix, latency needs, and budget.

View Kimi K2-Thinking-0905 profile

DeepSeek

Consider DeepSeek-V4-Flash-0423 when

your workload gives more weight to the higher listed coding score, the higher listed reasoning score, the higher overall index score. Compare those signals with your own task mix, latency needs, and budget.

View DeepSeek-V4-Flash-0423 profile

Compare a different pairRead editorial comparisons