Last verified
OPEN-WEIGHT FRONTIERFOUNDED 2023BEIJING · CHINA1T-PARAM MoELONG-CONTEXT KIMI

Kimi API Pricing

The lab behind Kimi — the long-context assistant — and the open-weight Kimi K2 models. Founded in 2023 by Yang Zhilin with Tsinghua schoolmates, Moonshot ships a 1-trillion-parameter open-weight MoE that rivals frontier US labs while staying downloadable on Hugging Face.

Current models
2
K2.6 + K2.5
Founded
2023
Beijing · Tsinghua founders
Context window
262K
Kimi K2 family
Open weights
1T MoE
K2 on Hugging Face
Cheapest current
$0.60/M
Kimi K2.5 input
Latest valuation
$20B
May 2026 round
§ 01 / LINEUP

The full roster.

Side-by-side →
FRONTIER · REASONING

Kimi K2.6

Moonshot's current flagship — native multimodal (text+image+video), thinking + non-thinking modes, agent capabilities.

Input
$0.95/M
Output
$4/M
262K ctx · vision
BALANCED · MID-TIER

Kimi K2.5

Previous-gen multimodal Kimi.

Input
$0.60/M
Output
$3/M
262K ctx · vision
PREVIEW · EARLY ACCESS

Moonshot V1 128K Vision Preview

Vision-input preview of Moonshot V1 128K.

Input
$2/M
Output
$5/M
131K ctx · vision
PREVIEW · EARLY ACCESS

Moonshot V1 32K Vision Preview

Vision-input preview of Moonshot V1 32K.

Input
$1/M
Output
$3/M
32K ctx · vision
PREVIEW · EARLY ACCESS

Moonshot V1 8K Vision Preview

Vision-input preview of Moonshot V1 8K.

Input
$0.20/M
Output
$2/M
8K ctx · vision
DEPRECATED · 2026-05-25

Kimi K2 (0905 Preview)

RETIRES 2026-05-25. Dated Sep-05 K2 preview snapshot.

Input
$0.60/M
Output
$3/M
262K ctx · text-only
DEPRECATED · 2026-05-25

Kimi K2 (0711 Preview)

RETIRES 2026-05-25. Dated Jul-11 K2 preview, smaller 131K context.

Input
$0.60/M
Output
$3/M
131K ctx · text-only
DEPRECATED · 2026-05-25

Kimi K2 Turbo Preview

RETIRES 2026-05-25. Turbo (low-latency) K2 preview.

Input
$1/M
Output
$8/M
262K ctx · text-only
DEPRECATED · 2026-05-25

Kimi K2 Thinking

RETIRES 2026-05-25. K2 reasoning-mode variant.

Input
$0.60/M
Output
$3/M
262K ctx · text-only
DEPRECATED · 2026-05-25

Kimi K2 Thinking Turbo

RETIRES 2026-05-25. Turbo reasoning variant.

Input
$1/M
Output
$8/M
262K ctx · text-only
LEGACY · STILL SUPPORTED

Moonshot V1 (128K)

Legacy Moonshot V1 line, 128K context.

Input
$2/M
Output
$5/M
131K ctx · text-only
LEGACY · STILL SUPPORTED

Moonshot V1 (32K)

Legacy Moonshot V1, 32K context tier.

Input
$1/M
Output
$3/M
32K ctx · text-only
LEGACY · STILL SUPPORTED

Moonshot V1 (8K)

Legacy Moonshot V1, smallest 8K context.

Input
$0.20/M
Output
$2/M
8K ctx · text-only
§ 02 / SHELF

All side-by-side.

Methodology →
Model Input /M Output /M Cached Context Max output Vision Tools Tier
Kimi K2.6 FLAGSHIP $0.95 $4.00 $0.16−83% 262K Frontier
Kimi K2.5 $0.6 $3.00 $0.1−83% 262K Mid
Moonshot V1 128K Vision Preview PREVIEW $2.00 $5.00 131K Preview
Moonshot V1 32K Vision Preview PREVIEW $1.00 $3.00 32K Preview
Moonshot V1 8K Vision Preview PREVIEW $0.2 $2.00 8K Preview
Kimi K2 (0905 Preview) DEPRECATED $0.6 $2.50 $0.15−75% 262K Deprecated
Kimi K2 (0711 Preview) DEPRECATED $0.6 $2.50 $0.15−75% 131K Deprecated
Kimi K2 Turbo Preview DEPRECATED $1.15 $8.00 $0.15−87% 262K Deprecated
Kimi K2 Thinking DEPRECATED $0.6 $2.50 $0.15−75% 262K Deprecated
Kimi K2 Thinking Turbo DEPRECATED $1.15 $8.00 $0.15−87% 262K Deprecated
Moonshot V1 (128K) LEGACY $2.00 $5.00 131K Legacy
Moonshot V1 (32K) LEGACY $1.00 $3.00 32K Legacy
Moonshot V1 (8K) LEGACY $0.2 $2.00 8K Legacy

The Kimi story so far.

Releases and funding from Moonshot's Kimi line. Sourced from Moonshot's GitHub/model cards, TechCrunch, and Wikipedia — verified at publication.

MAY 07 · 2026
$2B raised at a $20B valuation — led by Meituan's Long-Z, China's top-funded LLM startup
CORPORATE
APR 20 · 2026
Kimi K2.6 released — 1T-param open-weight MoE, native multimodal, four variants up to 300-agent swarms
RELEASE
MAY 25 · 2026
K2 preview series retires — dated K2 0711/0905, Turbo, and Thinking previews fold into K2.5 / K2.6
PRICING
JAN 27 · 2026
Kimi K2.5 released — multimodal upgrade adding a MoonViT vision encoder
RELEASE
JUL · 2025
Kimi K2 open-weighted — 1T-param MoE (32B active) trained with the MuonClip optimizer, released on Hugging Face
RELEASE
FEB · 2024
Alibaba-led $1B round — early backing at a $2.5B valuation
CORPORATE
OCT · 2023
Kimi assistant launched — famous for processing ~200,000 Chinese characters per conversation
RELEASE
§ 04 / ACCESS

Where to get it.

Methodology →
§ 04 / BEST FOR

Which Moonshot (Kimi) for what.

More scenarios →
If you need a frontier agentic multimodal model…
Kimi K2.6
Profile →
If you want cheaper multimodal at near-flagship quality…
Kimi K2.5
Profile →
If you must self-host an open-weight 1T MoE
Kimi K2 (HF)
Hugging Face →
If you process very long documents
Kimi K2.6 (262K)
Profile →
If you need a cheap short-context legacy tier…
Moonshot V1 8K
Profile →
If you need China data residency
Kimi Platform (China)
Kimi Platform →
§ 06 / BACKGROUND

The company behind it.

www.moonshot.ai →

Moonshot AI (月之暗面) was founded in March 2023 in Beijing by Yang Zhilin (a former Meta AI and Google Brain researcher) with Tsinghua University schoolmates Zhou Xinyu and Wu Yuxin. Its consumer assistant, Kimi, launched in October 2023 and built its reputation on very long context — handling roughly 200,000 Chinese characters per conversation, well ahead of peers at the time.

Moonshot's technical signature is the Kimi K2 line: a 1-trillion-parameter mixture-of-experts model with ~32B active parameters, trained with the lab's own MuonClip optimizer and released as open weights (Modified MIT) on Hugging Face. K2.5 (January 2026) added native multimodality via a MoonViT vision encoder, and K2.6 (April 2026) is the current open-weight flagship, shipping variants from quick chat up to large multi-agent swarms.

Funding has been rapid: an Alibaba-led $1B round in February 2024 (at a $2.5B valuation), rising to a $2B raise at a ~$20B valuation in May 2026 led by Meituan's Long-Z, with Tencent, HongShan, China Mobile, and others among its backers — making Moonshot one of China's top-funded LLM startups.

Pricing on the Kimi Platform is competitive, and the API offers automatic context caching that drops cache-hit input to a fraction of the base rate — meaningful for long-context workloads. The catalog has consolidated: the dated K2 preview models retire on May 25, 2026, leaving K2.6 and K2.5 as the current generation.

The trade-offs: Kimi is China-based, so the direct API stores data under Chinese rules — though the open weights let you self-host anywhere. Versus DeepSeek it competes directly on open-weight frontier quality; versus Qwen it offers fewer model sizes but a distinctive 1T-parameter open MoE and long-context pedigree.

§ 07 / COMPETITORS

Other frontier labs.

All providers →

Frequently asked.

Practical questions about Kimi K2, open weights, and long context.

Q · 01 Which Kimi model should I start with? +
For most workloads: Kimi K2.6 ($0.95/$4.00) — the open-weight flagship with multimodality, reasoning, and agentic tool use. For cheaper multimodal at near-flagship quality, use Kimi K2.5 ($0.60/$3.00). Both ship a 262K context. See the use case picker above.
Q · 02 Are Kimi models really open-weight? +
Yes. Kimi K2 is a 1-trillion-parameter MoE (≈32B active) released under a Modified MIT license on Hugging Face — fully downloadable and self-hostable. That makes Moonshot one of the few labs shipping a frontier-scale model as open weights.
Q · 03 What is Kimi's long-context heritage? +
Kimi made its name in 2023 by handling roughly 200,000 Chinese characters per conversation — far longer than rivals at the time. The current K2 models offer a 262K-token window, and the API's automatic context caching cuts cache-hit input to a fraction of the base rate, which matters for long documents.
Q · 04 Where is my data stored? +
The direct Kimi Platform API is China-based, so data is processed under Chinese data rules. US export-control and procurement questions apply for some buyers. Because K2 weights are open, the cleanest workaround for data control is to self-host them on your own US/EU infrastructure.
Q · 05 What happened to the older K2 preview models? +
The dated K2 previews — kimi-k2-0711, 0905, Turbo, and Thinking — retire on May 25, 2026. Moonshot's recommended replacements are K2.5 and K2.6, which are cheaper and stronger. New integrations should target the current generation.
Q · 06 How does Moonshot compare to DeepSeek and Qwen? +
Versus DeepSeek, Kimi K2 matches the open-weight frontier pitch but adds native multimodality (DeepSeek is text-only). Versus Qwen, Moonshot offers fewer sizes but a single very large 1T-parameter open MoE plus its long-context pedigree.