Model details

Moonshot Kimi K2.5

Moonshot Kimi K2.5 is designed as an all-in-one multimodal model, utilizing a Mixture-of-Experts architecture to handle both text and visual inputs. Its design intent centers on enabling sophisticated agent swarm orchestration, allowing enterprises to deploy autonomous agents that can pass actions between one another without requiring a central decision-making framework. With a massive scale of 1 trillion parameters, the model is built to excel in demanding coding and visual analysis projects, providing a robust foundation for developers looking to move beyond simple task execution toward interconnected, multi-agent workflows.

The model represents a significant advancement in open-source capabilities, demonstrating performance that competes with leading international models on key benchmarks. By supporting extensive context windows, it allows for deep integration into complex technical environments where large-scale data processing is required. Its architecture is particularly well-suited for forward-looking applications in industrial automation and visual software development, offering a flexible and powerful toolset for teams that need to scale their AI operations from initial concept to full production deployment.

Alibaba (China)kimi-k2.5kimi-k2

Open provider page Provider docs

Quick Info

Provider: Alibaba (China)
Model key: kimi-k2.5
Release date: Jan 27, 2026
Last updated: Jan 27, 2026
Knowledge cutoff: 2025-01
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.574
Output token cost: $2.411

Limits

Output tokens: 32,768 tokens
Context window: 262,144 tokens

Transparent token rates

Compare Moonshot Kimi K2.5 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Moonshot Kimi K2.5

Alibaba (China)

Coverage

A mid-2026 open-source LLM roundup on tech-insider.org ranks the five open-weight models that matter most in mid-2026: DeepSeek V4-Pro, Moonshot's Kimi K2.6, Zhipu's GLM-4.6, Alibaba's Qwen3-235B-A22B, and Meta's Llama 4 Maverick. The article notes Kimi K2.6 is positioned as the model to beat for long-running agentic a Note: this candidate references the Kimi K2 family successor (K2.6) rather than K2.5 specifically, but provides useful positioning context for the Kimi K2 line within the broader open-weight landscape, including how Moonshot's releases compare against DeepSeek, Zhipu, Alibaba, and Meta offerings.

Alibaba (China)

CoverageBenchmark

BenchLM.ai's provider page for Moonshot AI documents the full model lineup as of September 2026, showing Kimi K2.5 with 256K context window, $0.60/$3.00 per million tokens (input/output), 82 tokens/second speed, ~38.77s latency, and a public score of 55.62. The page marks K2.5 as "Superseded" status, tracking 7 models Newer Kimi K2 family releases are also tracked on the same provider: Kimi K2.6 (released April 20, 2026, now superseded), Kimi K2.7 Code (released June 12, 2026, current), and Kimi K3 (released July 16, 2026, current flagship with 1.05M context at $3.00/$15.00 per million tokens). The data provides concrete pricing and