Requesty
Google LLC (Vertex AI) logo
Google LLC (Vertex AI)

kimi-k2

Kimi K2 Thinking is an open-source model that operates as a "thinking agent," reasoning step-by-step while using tools to achieve state-of-the-art performance on various benchmarks. It is capable of executing up to 200-300 sequential tool calls without human intervention, allowing it to solve complex problems across a wide range of tasks. The model uses Quantization-Aware Training (QAT) to support INT4 inference, which provides a roughly 2x improvement in generation speed.

👁Vision🧠Reasoning🔧Tool calling⚡Caching
Try kimi-k2CompareDocs

Specifications

Context window262K tokens
Max output262K tokens
API typechat
AddedApr 16, 2026
Model IDvertex/kimi-k2
Data retentionNo
Used for trainingNo
Provider location🇺🇸 US / 🇪🇺 EU
Privacy policyVertex AI Data Governance→

Benchmarks

Released 2025-07
SWE-Bench Verifiedcoding
65.8%

Resolving real GitHub issues from 12 popular Python repositories.

GPQA Diamondreasoning
70.0%

Graduate-level physics, chemistry & biology questions designed to resist Googling.

MMLU Proknowledge
82.3%

Massive Multitask Language Understanding across 57 academic subjects.

Scores are sourced from official model cards, Artificial Analysis, and public leaderboards. Benchmarks measure specific skills and do not capture every aspect of model quality — always test on your own workload.

Pricing

Input / 1M
$0.60
Output / 1M
$2.50
Cache write / 1M
$2.50
Cache read / 1M
$0.06
Estimated cost
100K input + 10K output$0.0850
1M input + 100K output$0.85
10M input + 1M output$8.50

Requesty charges exactly what the upstream provider charges — no markup, no per-request fees. Prompt caching and smart routing can reduce effective cost by 30-80%.

Quickstart

Drop-in compatible with the OpenAI SDK. Change the base URL, swap in your Requesty API key, and set the model to vertex/kimi-k2.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_REQUESTY_API_KEY",
    base_url="https://router.requesty.ai/v1",
)

response = client.chat.completions.create(
    model="vertex/kimi-k2",
    messages=[
        {"role": "user", "content": "Explain quantum computing in one paragraph."},
    ],
)

print(response.choices[0].message.content)

Other Google LLC (Vertex AI) models

gemini-3.5-flash
@eu
1.0M ctx$1.50/1M
gemini-3.5-flash
1.0M ctx$1.50/1M
gemini-3.5-flash
@us
1.0M ctx$1.50/1M
gemini-3.1-flash-lite
@us
1.0M ctx$0.25/1M
gemini-3.1-flash-lite
1.0M ctx$0.25/1M
gemini-3.1-flash-lite
@eu
1.0M ctx$0.25/1M

Frequently asked questions

How much does kimi-k2 cost?
kimi-k2 is priced at $0.60 per million input tokens and $2.50 per million output tokens when accessed via Requesty. Prompt caching is supported, which can cut effective input cost by up to 90% on repeated context. Requesty charges exactly what the upstream provider charges — we don't add markup.
What is the context window of kimi-k2?
kimi-k2 has a context window of 262K tokens, with a maximum output of 262K tokens per response. That's roughly 350 words of input you can fit in a single prompt.
How does kimi-k2 perform on benchmarks?
kimi-k2 scores 89.9% on HumanEval, 89.2% on MATH, 82.3% on MMLU Pro. See the full benchmark chart above for results across MMLU Pro, GPQA Diamond, SWE-Bench Verified, HumanEval, MATH, AIME, MMMU, and LiveBench.
What can kimi-k2 do?
kimi-k2 supports vision input, tool calling, extended reasoning, prompt caching. You can call it through any OpenAI-compatible client by pointing base_url to Requesty.
How do I use kimi-k2 with the OpenAI SDK?
Install the OpenAI SDK, set base_url to "https://router.requesty.ai/v1", set your API key to your Requesty key, and set the model to "vertex/kimi-k2". The Quickstart above shows Python, JavaScript and cURL snippets.

Access kimi-k2 through Requesty

One API key, 400+ models, OpenAI-compatible. No markup on provider prices, automatic failover, and smart caching built-in.

Get started freeAll Google LLC (Vertex AI) models