Moonshot AI
Updated May 2026
Kimi
The long-context and agentic-work specialist
Moonshot AI, based in Beijing, built its reputation on extreme long-context — early Kimi handled 2M characters before any US model came close. Kimi K2 in 2025 moved into agentic coding, and K2.6 in April 2026 can coordinate swarms of 300 sub-agents across 4000 steps autonomously. Open weights are released on HuggingFace, making Kimi the go-to for teams building autonomous software engineers.
Variants
4
Best at
best-in-class agent orchestration
Max context
2M
tokens
Pricing
Free
$0
per kimi.com chat
K2.6 API
$0.60 in / $2.50 out
per million tokens
Variants
Sort the table by context window or cost to find the right variant. Click any version below for a battle card with ranks, pricing notes, and official links.
| Modalities | ||||
|---|---|---|---|---|
Kimi K1 Long-Context kimi-k1 | 2M | varies / varies | 2024 | text |
Kimi K2.6 kimi-k2-6 | 256K | varies / varies | 2026 | textvisionvideo |
Kimi K2.5 kimi-k2-5 | 256K | varies / varies | 2025 | textvision |
Kimi K2 kimi-k2 | 128K | varies / varies | 2025 | text |
Battle card
Context rank
#2
within Kimi
Capability rank
#1
modalities + reasoning
Weights
Open
self-hostable if licensed
Best fights to pick
- long-horizon coding agents
- autonomous software engineering
- agent-swarm orchestration
Rankings are Tendril directory ranks, computed from the model data shown here. Public benchmark leaderboards change often, so official docs and current benchmark pages should be checked before buying or deploying.
Learn
Lessons about this model
Structured lessons that cover Kimi directly or put it in context alongside its rivals.
Check yourself
Quizzes
Short, mixed-difficulty quiz sets on Kimi and its model family.
Other Players: MiniMax, Cohere, Amazon, 01.AI, and More
5 questions
The broader landscape of model providers.
Start quiz →Hands-on
Try these prompts
Ready-made prompts that show Kimi at its best. Use them in your own AI workspace, then compare the output with what you learned in Tendril.
Kimi on a whole codebase
CreatorsKimi K2's long context is its flagship feature — feed it a repo and ask architectural questions.
I'm pasting my entire Next.js app below (about 50 files). Find places where we duplicate form-validation logic, and propose a single shared utility that could replace all of them. Show the refactor plan, not the full diff yet. [Paste codebase]
Kimi Research Mode
CreatorsKick off an autonomous research pipeline on a niche question.
Research small-scale aquaponics systems for apartment balconies. I need: typical startup cost, ongoing water/energy use, what fish species actually thrive in this constraint, and the top 3 mistakes beginners make. Cite sources.
Long-document Q&A
BuildersDrop a book-length PDF in and quiz Kimi on specific pages.
Using only the attached 400-page textbook, answer: on which page(s) does the author argue that cognitive load theory contradicts active learning, and what's the exact argument? Quote the key sentence.
Print & keep
Printable reference
One-page summaries and flowcharts — great for desks, classrooms, or study sessions.
Go deeper
Official resources
Straight from the lab — docs, API references, and the chat surfaces you can try today.
Strengths
- best-in-class agent orchestration
- open weights at frontier coding
- competitive pricing
Limits
- China-based, data-residency concerns abroad
- documentation mostly in Chinese
- smaller international ecosystem