Kimi K2.5

by Moonshot AI

+
+
+
+
About

Kimi K2.5, released by Moonshot AI in January 2026, is an updated Mixture-of-Experts large language model with 1 trillion total parameters and 32 billion active parameters. It builds on Kimi K2 with improved coding performance across multiple languages and an expanded context window. Kimi K2.5 targets agentic development workflows, polyglot code generation, and open-source deployments requiring large-scale MoE reasoning.

+
+
+
+
Pricing Range
Input (per 1M)$1.50 -$1.50
Output (per 1M)$6.00 -$6.00
Providers1
+
+
+
+
Timeline
ReleasedJan 1, 2026
+
+
+
+
License & Family
License
MIT
Performance Overview
Performance metrics and category breakdown

Overall Performance

5 benchmarks
Average Score
49.1%
Best Score
76.8%
High Performers (80%+)
0

Performance Metrics

Max Context Window
264.2K

Top Categories

Agents
63.3%
Coding
52.6%
Reasoning
24.4%
+
+
+
+
All Benchmark Results for Kimi K2.5
Complete list of benchmark scores with detailed information
SWE Bench Verified
Coding
76.80
76.8%
Unverified
OSWorld
Agents
63.30
63.3%
Unverified
Terminal Bench 2.0
Coding
43.20
43.2%
Unverified
SWE-rebench
Coding
37.90
37.9%
Unverified
Humanity's Last Exam
Reasoning
24.37
24.4%
Unverified