Moonshot Kimi K2.5 Pricing, Benchmarks & Channels

Moonshot Kimi K2.5 is a 2026-01-27 model from Moonshot AI (Kimi) featuring a 262K context window and extended reasoning capabilities. Lowest rate on SiliconFlow (CN) ($0.45 in / $2.25 out per 1M tokens).

⚖️ Compare in Radar Workbench 🧮 Monthly Cost Simulator
Input Price / 1M
$0.574
Output Price / 1M
$3.011
Cache Read / 1M
$0.1148
Context Window
262K
Generation Speed
58 TPS
Access Grade
A · CN Direct

F.A.R.E. Four-Dimension Score Breakdown

A · Accurate (Capability) 76/100 · Elo 1451 F · Fast (Speed & Latency) 58 Token/s · TTFT 5290ms R · Reliable (Stability & Access) Grade A · CN Domestic / CN Direct E · Economical (Cost) $0.574 in · $3.011 out
Scenario Capability Breakdown (0–100):
💻 Coding
76/100
✍️ Writing
72/100
📚 Long-Doc RAG
77/100
👁️ Vision OCR
80/100
F.A.R.E. CAPABILITY RADAR
A · Accurate F · Fast E · Economical R · Reachable

Moonshot Kimi K2.5 Multi-Platform Price Comparison (6 Verified Channels)

All prices are in USD per 1M tokens. Domestic developers can prioritize channels marked "Direct in CN" for compliant access and local billing.

Platform / Gateway Network Access Input ($/1M) Output ($/1M) Cache Hit ($/1M) Billing & Notes Link
SiliconFlow (CN) Lowest Price Direct in CN $0.45 $2.25 $0.07 Direct CN access · RMB billing Visit Official →
DeepInfra Global Network $0.45 $2.25 $0.07 Global cloud billing Visit Official →
OpenRouter Direct in CN $0.5 $2.5 $0.15 Direct CN access · RMB billing Visit Official →
Alibaba Cloud Bailian (CN) Direct in CN $0.574 $3.011 $0.1148 Direct CN access · RMB billing Visit Official →
AWS Bedrock Global Network $0.6 $3 $0.15 Global cloud billing Visit Official →
302.AI Gateway Direct in CN $0.627 $3.3 $0.1568 Direct CN access · RMB billing Visit Official →

✅ Core Strengths & Ideal Workloads

  • Direct access in mainland China with RMB enterprise billing
  • Open weights available for private self-hosting (~74GB VRAM at INT4)

⚠️ Limitations & Practical Caveats

  • Direct domestic API and global endpoints; Kimi K3.1 Pro leads open-weight coding and 1M+ long-context tasks.
  • Extended thinking generates internal reasoning tokens (~1.8x output multiplier)