MoonshotAI: Kimi K2 0905

Text input Text output
Author's Description

Kimi K2 0905 is the September update of [Kimi K2 0711](moonshotai/kimi-k2). It is a large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI, featuring 1 trillion total parameters with 32 billion active per forward pass. It supports long-context inference up to 256k tokens, extended from the previous 128k. This update improves agentic coding with higher accuracy and better generalization across scaffolds, and enhances frontend coding with more aesthetic and functional outputs for web, 3D, and related tasks. Kimi K2 is optimized for agentic capabilities, including advanced tool use, reasoning, and code synthesis. It excels across coding (LiveCodeBench, SWE-bench), reasoning (ZebraLogic, GPQA), and tool-use (Tau2, AceBench) benchmarks. The model is trained with a novel stack incorporating the MuonClip optimizer for stable large-scale MoE training.

Key Specifications
Cost
$$$$
Context
262K
Parameters
32B (Rumoured)
Released
Sep 04, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Stop Top P Structured Outputs Response Format Frequency Penalty Tool Choice Max Tokens Tools Presence Penalty Temperature
Features

This model supports the following features:

Response Format Tools Structured Outputs
Performance Summary

MoonshotAI's Kimi K2 0905 demonstrates moderate speed performance, ranking in the 23rd percentile across six benchmarks, indicating it is not among the fastest models available. Its pricing is also moderate, placing it in the 35th percentile, suggesting competitive but not exceptionally low costs. A standout feature is its exceptional reliability, achieving a 100% success rate across all benchmarks, signifying minimal technical failures and consistent response delivery. In terms of benchmark performance, Kimi K2 0905 excels in Instruction Following (85th percentile accuracy) and Email Classification (84th percentile accuracy), showcasing strong capabilities in precise directive execution and categorization. Its Coding (Baseline) performance is solid at 87.0% accuracy (68th percentile), aligning with its described improvements in agentic and frontend coding. However, the model exhibits notable weaknesses in Reasoning (47th percentile accuracy), Ethics (21st percentile accuracy), and General Knowledge (24th percentile accuracy), where its performance falls below average compared to other models. While its coding and instruction following are strengths, its lower scores in critical thinking and broad knowledge areas suggest room for improvement in these domains.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.6
Completion $2.5
Input Cache Read $0.15

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Moonshot AI
Moonshot AI | moonshotai/kimi-k2-0905 262K $0.6 / 1M tokens $2.5 / 1M tokens
Moonshot AI
Moonshot AI | moonshotai/kimi-k2-0905 262K $2.4 / 1M tokens $10 / 1M tokens
Together
Together | moonshotai/kimi-k2-0905 262K $1 / 1M tokens $3 / 1M tokens
Fireworks
Fireworks | moonshotai/kimi-k2-0905 262K $0.6 / 1M tokens $2.5 / 1M tokens
Groq
Groq | moonshotai/kimi-k2-0905 262K $1 / 1M tokens $3 / 1M tokens
Novita
Novita | moonshotai/kimi-k2-0905 262K $0.6 / 1M tokens $2.5 / 1M tokens
Chutes
Chutes | moonshotai/kimi-k2-0905 262K $0.296 / 1M tokens $1.19 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration
Other Models by moonshotai