Author's Description
Kimi K2 Thinking is Moonshot AI’s most advanced open reasoning model to date, extending the K2 series into agentic, long-horizon reasoning. Built on the trillion-parameter Mixture-of-Experts (MoE) architecture introduced in Kimi K2, it activates 32 billion parameters per forward pass and supports 256 k-token context windows. The model is optimized for persistent step-by-step thought, dynamic tool invocation, and complex reasoning workflows that span hundreds of turns. It interleaves step-by-step reasoning with tool use, enabling autonomous research, coding, and writing that can persist for hundreds of sequential actions without drift. It sets new open-source benchmarks on HLE, BrowseComp, SWE-Multilingual, and LiveCodeBench, while maintaining stable multi-agent behavior through 200–300 tool calls. Built on a large-scale MoE architecture with MuonClip optimization, it combines strong reasoning depth with high inference efficiency for demanding agentic and analytical tasks.
Key Specifications
Supported Parameters
This model supports the following parameters:
Features
This model supports the following features:
Performance Summary
MoonshotAI's Kimi K2 Thinking model, an advanced open reasoning model, demonstrates exceptional reliability with a 100% success rate across all benchmarks, indicating consistent and stable operation. However, it tends to have longer response times, ranking in the 1st percentile for speed, and is positioned at premium pricing levels, ranking in the 13th percentile for cost. Despite its slower speed and higher cost, Kimi K2 Thinking excels in several critical areas. It achieves near-perfect accuracy in Coding (97.6%, 99th percentile) and Mathematics (96.8%, 99th percentile), placing it among the top performers in these demanding categories. Its Instruction Following capability is also outstanding at 88.5% accuracy (96th percentile), making it the most accurate model at its price point for this task. The model also achieves perfect accuracy in Ethics (100%), showcasing strong moral reasoning. While its General Knowledge (99.5%, 72nd percentile) and Email Classification (99.0%, 70th percentile) are solid, its Hallucinations accuracy (90.0%, 37th percentile) suggests some room for improvement in acknowledging uncertainty. Its Reasoning performance is respectable at 83.7% (73rd percentile). The model's strength lies in its ability to handle complex, multi-step tasks, leveraging its agentic architecture for autonomous research, coding, and writing over hundreds of turns.
Model Pricing
Current Pricing
| Feature | Price (per 1M tokens) |
|---|---|
| Prompt | $0.4 |
| Completion | $1.75 |
Price History
Available Endpoints
| Provider | Endpoint Name | Context Length | Pricing (Input) | Pricing (Output) |
|---|---|---|---|---|
|
Moonshot AI
|
Moonshot AI | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.4 / 1M tokens | $1.75 / 1M tokens |
|
Moonshot AI
|
Moonshot AI | moonshotai/kimi-k2-thinking-20251106 | 262K | $1.15 / 1M tokens | $8 / 1M tokens |
|
Parasail
|
Parasail | moonshotai/kimi-k2-thinking-20251106 | 128K | $0.4 / 1M tokens | $1.75 / 1M tokens |
|
Moonshot AI
|
Moonshot AI | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.6 / 1M tokens | $2.5 / 1M tokens |
|
Novita
|
Novita | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.48 / 1M tokens | $2 / 1M tokens |
|
Parasail
|
Parasail | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.5 / 1M tokens | $2.5 / 1M tokens |
|
Chutes
|
Chutes | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.4 / 1M tokens | $1.75 / 1M tokens |
|
GMICloud
|
GMICloud | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.8 / 1M tokens | $1.2 / 1M tokens |
|
Fireworks
|
Fireworks | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.4 / 1M tokens | $1.75 / 1M tokens |
|
Together
|
Together | moonshotai/kimi-k2-thinking-20251106 | 262K | $1.2 / 1M tokens | $4 / 1M tokens |
|
DeepInfra
|
DeepInfra | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.4 / 1M tokens | $1.75 / 1M tokens |
|
AtlasCloud
|
AtlasCloud | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.6 / 1M tokens | $2.5 / 1M tokens |
|
Google
|
Google | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.6 / 1M tokens | $2.5 / 1M tokens |
|
SiliconFlow
|
SiliconFlow | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.55 / 1M tokens | $2.5 / 1M tokens |
|
BaseTen
|
BaseTen | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.6 / 1M tokens | $2.5 / 1M tokens |
|
Nebius
|
Nebius | moonshotai/kimi-k2-thinking-20251106 | 262K | $0.6 / 1M tokens | $2.5 / 1M tokens |
|
DeepInfra
|
DeepInfra | moonshotai/kimi-k2-thinking-20251106 | 131K | $0.47 / 1M tokens | $2 / 1M tokens |
Benchmark Results
| Benchmark | Category | Reasoning | Strategy | Free | Executions | Accuracy | Cost | Duration |
|---|
Other Models by moonshotai
|
|
Released | Params | Context |
|
Speed | Ability | Cost |
|---|---|---|---|---|---|---|---|
| MoonshotAI: Kimi Linear 48B A3B Instruct Unavailable | Nov 07, 2025 | 48B | 1M |
Text input
Text output
|
★★★★ | ★★ | $$$ |
| MoonshotAI: Kimi Linear 48B A3B Instruct Unavailable | Nov 07, 2025 | 48B | 1M |
Text input
Text output
|
— | ★ | $$$$ |
| MoonshotAI: Kimi K2 0905 | Sep 04, 2025 | ~32B | 262K |
Text input
Text output
|
★★ | ★★★ | $$$$ |
| MoonshotAI: Kimi K2 0905 (exacto) | Sep 04, 2025 | ~1T | 262K |
Text input
Text output
|
— | — | $$$$$ |
| MoonshotAI: Kimi K2 0711 | Jul 11, 2025 | ~1T | 131K |
Text input
Text output
|
★★★★ | ★★★★★ | $$ |
| MoonshotAI: Kimi Dev 72B | Jun 16, 2025 | 72B | 131K |
Text input
Text output
|
★ | ★★ | $$$$ |
| MoonshotAI: Kimi VL A3B Thinking Unavailable | Apr 10, 2025 | 3B | 131K |
Text input
Image input
Text output
|
★ | ★ | $$$ |