# Kimi K2

Kimi K2 is an AI model released by Moonshot AI on Jul 11 2025. It has 1T parameters, a 128k context window and open weight. Tracked results include 10% on BullshitBench v2, 65.8% on SWE-Bench Verified and 75.1% on GPQA Diamond.

## Facts

| Field | Value |
| --- | --- |
| Model | Kimi K2 |
| Developer | Moonshot AI |
| Release date | Friday, Jul 11 2025 |
| Licensing | Open Weight |
| Parameters | 1T |
| Context window | 128k |

## Tracked benchmark scores

| Benchmark | Score | Source | What it measures |
| --- | --- | --- | --- |
| BullshitBench v2 | 10% | [BullshitBench](https://github.com/petergpt/bullshit-benchmark) | Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. |
| SWE-Bench Verified | 65.8% | Lab | Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. |
| Humanity's Last Exam (with tools) | 4.7% | [BenchLM](https://benchlm.ai), retrieved 2026-10-12 | Humanity's Last Exam — extremely hard expert questions across many subjects. “With tools” means the AI is allowed to search the web or run code while answering. Higher is better. |
| GPQA Diamond | 75.1% | Lab | Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. |

## About Kimi K2

Kimi K2 was released on July 11, 2025 as Moonshot AI's mixture-of-experts model for general and agentic tasks. It had one trillion total parameters and about 32 billion active per token, with base and instruction-tuned weights available. Moonshot emphasized tool use, coding and carrying out multi-step requests, extending the Kimi line beyond a hosted conversational product.

The training report introduced the MuonClip optimizer and described large-scale training for a sparse model. The initial instruction checkpoint supported a 128,000-token context window. Its tool-use focus made it relevant to agent systems that needed to choose functions, interpret results and continue a task across multiple interactions.

The weights were published under a modified MIT License with an additional attribution requirement for very large deployments. That provided a self-hosting route as well as hosted API access, while retaining specific release conditions. Later K2 checkpoints expanded context and reasoning capabilities, so their specifications and benchmarks should be kept separate from this original July release.

## Questions and answers

### When was Kimi K2 released?

Kimi K2 was released by Moonshot AI on Friday, Jul 11 2025.

### Who made Kimi K2?

Kimi K2 was built by Moonshot AI. Chinese AI company building the Kimi family of models. Founded in 2023, known for long-context and efficient MoE architectures.

### What benchmark scores did Kimi K2 get?

Kimi K2 reports 4 tracked benchmark scores — BullshitBench v2: 10%; SWE-Bench Verified: 65.8%; Humanity's Last Exam (with tools): 4.7%; GPQA Diamond: 75.1%. Tracked scores may come from lab reports or independent benchmarks; source details accompany the benchmark data.

### What is the context window of Kimi K2?

Kimi K2 has a context window of 128k. Providers may specify separate input and output token limits.

### How many parameters does Kimi K2 have?

Kimi K2 is reported at 1T parameters.

### Is Kimi K2 open source?

Partly. Kimi K2 is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.

### What came before and after Kimi K2?

Moonshot AI's previous tracked release was Kimi-Dev on Jun 17 2025, 24 days earlier. It was followed by Kimi K2 (0905) on Sep 5 2025.


---

Canonical page: https://aireleasetracker.com/model/moonshot/kimi-k2
Site index: https://aireleasetracker.com/llms.txt
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.