Kimi K2.5

Extracted Attributes

• License

  MIT License

• Developer

  Moonshot AI

• Release Date

  2025-07-01

• Training Data

  15 trillion mixed visual and text tokens

• Context Window

  262,000 tokens

• Pricing (Input)

  $0.60 per 1 million tokens

• Input Modalities

  Text, Image, Video

• Pricing (Output)

  $3.00 per 1 million tokens

• Total Parameters

  1 trillion

• Active Parameters

  32 billion

• Model Architecture

  Mixture-of-Experts (MoE)

Timeline

• First version of Kimi released by Moonshot AI, supporting 128,000 tokens of context. (Source: Wikipedia)

  2023-01-01

• Release of Kimi K2 and Kimi K2.5 as open-weight models with strong coding and agentic capabilities. (Source: Wikipedia)

  2025-07-01

Kimi (chatbot)

Kimi is an artificial intelligence (AI) chatbot and series of large language models developed by Chinese company Moonshot AI. Its first version, released in 2023, was known for supporting up to 128,000 tokens of context. Kimi K2, an open-weight model released in July 2025, showed strong performances on coding benchmarks.

Web Search Results

• Kimi K2.5: Visual Agentic Intelligence | Technical Report

  All reported scores of coding tasks are averaged over 5 independent runs. [...] Note: The coordinates (row, column) are 0-indexed from the top-left. The actual path consists of 113,557 connected pixels forming a continuous route from start to finish without any shortcuts possible (as guaranteed by BFS optimality). K2.5 excels in real-world software engineering tasks. We evaluate it using Kimi Code Bench, our internal coding benchmark covering diverse end-to-end tasks — from building to debugging, refactoring, testing, and scripting — across multiple programming languages. On this benchmark, K2.5 shows consistent and meaningful improvements over K2 across task types. [...] To try out K2.5's agentic coding capabilities, K2.5 Agent offers a set of preconfigured tools for immediate, hands-on experiences. For software engineering use cases, we recommend pairing Kimi K2.5 with our new coding product, Kimi Code. Kimi Code works in your terminal and can be integrated with various IDEs including VSCode, Cursor, Zed, etc. Kimi Code is open-sourced and supports images and videos as inputs. It also automatically discovers and migrates existing skills and MCPs into your working environment in Kimi Code.

• Kimi K2.5 Is Live on Fireworks: Vibe Coding, Agents, and Full ...

  Kimi K2.5 Is Live on Fireworks with Full-Parameter RL tuning. Try It Today. Log InGet Started BlogKimi K2p5 # Kimi K2.5 is Live on Fireworks: Vibe Coding, Agents, and Full-Parameter RFT Kimi K2.5 is Moonshot AI’s flagship agentic model and a new SOTA open model. It unifies vision and text, thinking and non-thinking modes, and multi-agent execution into one model. We are launching Day-0 support for Kimi K2.5. Fireworks offers the fastest endpoint for all Kimi K2 series models as well as fine tuning for Kimi K2 models. Additionally, we now offer a full parameter RL tuning private preview for Kimi K2.5, enabling application builders to fine tune the SOTA OSS VLM model for use cases like vibe coding and agentic workflows. Sign up for the full parameter RL tuning waitlist here.

• Kimi K2.5 in 2026: The Ultimate Guide to Open-Source Visual ...

  Multiple Access Methods: Available via Kimi.com, API ($0.60/M input, $3/M output), Kimi Code CLI, and direct model weights on HuggingFace [...] Open-Source Breakthrough: Kimi K2.5 is a 1 trillion parameter MoE model (32B active) with MIT license, representing the most powerful open-weight multimodal model available Revolutionary Agent Swarm: Self-directs up to 100 sub-agents executing 1,500+ parallel tool calls, achieving 4.5× speed improvement through Parallel-Agent Reinforcement Learning (PARL) Native Multimodal Architecture: Built from ground-up with 15T mixed visual and text tokens, delivering SOTA coding with vision and autonomous visual debugging Competitive Performance: Matches or exceeds GPT-5.2, Claude 4.5 Opus, and Gemini 3 Pro across multiple benchmarks while remaining fully accessible [...] ### Kimi Code Bench Performance On the internal Kimi Code Bench (covering building, debugging, refactoring, testing, and scripting across multiple languages), K2.5 shows consistent improvements over K2 across all task types. > 💡 Pro Tip > > For software engineering use cases, pair Kimi K2.5 with Kimi Code—an open-source CLI tool that integrates with VSCode, Cursor, Zed, and other IDEs. It supports images and videos as inputs and automatically discovers existing skills and MCPs. ## Performance Benchmarks Comparison ### Reasoning & Knowledge Benchmarks

• Kimi K2.5: The GREATEST Opensource AI Model That Beats Opus ...

  deeper processing. Then is the Kim K 2.5 agent for agentic workflows. And lastly, the Kimmy K 2.5 agent swarm, which we just talked about. What's especially impressive is that the agent swarm is fully self-created and orchestrated by the model itself with no predefined agents or manual workflow setup. In terms of benchmarks, EK 2.5 is evaluated across a wide range of agentic, coding, vision, math, document, and video benchmarks, including HLE, full browser comp, Swaybench, and so many others. On top of that, Kim K2.5 excels at realworld software engineering tasks from building and debugging to refactoring, testing, and scripting across multiple programming languages. It also is particularly strong in front-end development. And honestly, some of the generations that I have seen clearly [...] 42 comments ### Transcript: [...] output and this is the capability of the 2.5 with all these different modes like the Asia mode in this case that is executing all of these different tasks. In regards to pricing, the Kimik K 2.5 is extremely aggressive. It's listed at 60 cents per million input tokens and just 10 cents per million with a cash hit. It also has $3 per 1 million output tokens listed. But what's wild is that this comes in at roughly 10% of the cost of Opus at similar similar performance levels and about 20% the cost of Claude 4.5 sonnet all while supporting a massive 262k token context window. But guys, I want to step back and make a note that this is an open-source model that is comparable to models like Gemini as well as something like Opus 4.5, which is just incredible. We have an open- source model with

• Kimi K2.5 Technical Report [pdf] - Hacker News

  Image 89 Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact Search: [...] I'm not very familiar with this model series but tried it when prototyping a potential 'Vibesbench' (my eval) conversation I tried it. Here's Kimi on a Robocop scene kimi k2 thinking turbo "He’s not being poetic; he’s weaponizing nostalgia. Quoting Casablanca—the ultimate “beautiful sacrifice” romance—to taunt Morton about corporate backstabbing is pure executive sleaze. It’s the same move as his mahogany desk and flag pin: draping sentimental Americana over naked power. He’s essentially saying, “Isn’t it romantic, Bob? You fought the good fight, now you die.” " grok 4.1 "So when Dick Jones smirks and says, “Time goes by, it’s the same old story… the fight for love and glory, huh, Bob?” that’s pure Ed Neumeier flexing his love of old Hollywood while simultaneously mocking how these [...] 37) That is a very good point and I would love to do it, but I built this machine in a desktop case and the motherboard has seven slots. I did a custom water cooling manifold just to make it work with all the cards. I'm trying to figure out how to add another card on a riser hanging off a slimsas port, or maybe I could turn the bottom slot into two vertical slots.. the case (fractal meshify 2 xl) has room for a vertical mounted card that wouldn't interfere with the others, but I'd need to make a custom riser with two slots on it to make it work. I dunno, it's possible! I also have an RTX Pro 6000 Blackwell and an RTX 5000 Ada.. I'd be better off pulling all the A7000s and throwing both of those cards in this machine, but then I wouldn't have anything for my desktop. Decisions, decisions!