NativeAIHub

K2.6 / K2.5 Architecture, Benchmarks, and Operating Modes

1 min read

0T params

Total model parameters

0B active

Active per token (MoE)

0K tokens

Context window

0

Expert modules

K2.6 ScoreK2.5 ScoreContext
SWE-Bench Verified80.276.8Competitive with Claude Opus 4.6
SWE-Bench Pro58.650.7Leads GPT-5.4 and Claude Opus 4.6
SWE-Bench Multilingual76.773.0Near Claude Opus 4.6 level
AIME 202696.495.8Advanced math competition problems
GPQA Diamond90.587.6Graduate level science questions
HLE-Full (with tools)54.050.2Leads GPT-5.4 and Claude Opus 4.6
LiveCodeBench (v6)89.685.0Real time coding evaluation
Terminal-Bench 2.066.750.8Shell and terminal task completion
BrowseComp83.274.9Long-horizon web browsing agents
SpeedBest For
InstantFastestQuick questions, simple tasks, casual conversation
ThinkingMediumMath, logic, complex analysis, coding problems
AgentSlowerResearch, multi-step tasks, file processing
Agent Swarm (beta)VariableDeep research, large codebase analysis, comprehensive investigations

Why MoE matters for cost

With only 32 billion parameters active per token out of 1 trillion total, K2.6 and K2.5 achieve frontier performance while using a fraction of the compute per inference. This is why API pricing can be $0.95/$4.00 (K2.6) or $0.60/$3.00 (K2.5) per million tokens while matching models that cost 5x to 25x more.

K2.5 in third party tools: the Cursor controversy

In early 2026, it was revealed that Cursor uses Kimi K2.5 in its Composer 2 feature. This sparked debate about model attribution and transparency, with users questioning whether AI coding tools should clearly disclose which underlying models power their features. On the positive side, it validates K2.5's coding capabilities at the frontier level, since a leading AI code editor chose it for a core feature.