How to Use Kimi K3 This Week: App, Free Try, API & Code
Kimi K3 hub: Specs, pricing, API id → /kimi-k3. Timeline → /kimi-k3-status. Free options → how to try K3 without a big bill.
Your timeline is still full of Kimi K3. One post says it’s free. The next says new subscriptions are paused. Someone else is already arguing about Ollama and Hugging Face.
You’re left with a simpler problem:
What should you actually open today—and what can wait until next week?
Short answer (read this first)
Pick one door, not five tabs:
| You want… | Open this first |
|---|---|
| Just try K3 in a chat/app | kimi.com / Kimi app — start free via the Moon Landing Plan invite when you can claim rewards |
| Checkout is blocked / “new subscriptions paused” | Don’t force a card. Stay on free/credit limits, or use the paid API if you build software. Details: subscription pause guide |
| Stable automation / app integration | Pay-as-you-go kimi-k3 on platform.kimi.ai (~$0.30 cache-hit input · $3 input · $15 output per 1M tokens) |
| Day-to-day coding in an IDE/CLI | Prefer K2.7 Code first for routine loops; use K3 when the job is hard and long. Hub: /kimi-k27 · /kimi-code |
| “I’ll self-host the full model this weekend” | Almost certainly not yet. Full weights are promised by July 27, 2026—verify the real repo when it lands. Open-weights reality check |
This site is independent of Moonshot. Product capacity and checkout can change overnight—re-check kimi.com, platform.kimi.ai, and @Kimi_Moonshot before you plan a team rollout.

Source: Official @Kimi_Moonshot K3 launch media, July 16, 2026; same charts as the Kimi K3 blog.
Door A — Try K3 in the product (app / web)
If your goal is “I want to feel K3,” skip the API dashboard for now.
- Open kimi.com or the Kimi mobile app.
- Prefer the official Moon Landing Plan path so free signup can include membership-credit rewards: Sign up free and claim rewards. The prize ladder is official marketing (including up to 1 year of membership-equivalent value)—not a guarantee you’ll pull the top prize.
- Ask for something real: a multi-file refactor plan, a long research pack, a UI from a screenshot, a spreadsheet-style reasoning task—not “hi, who are you?”
- Watch quotas and context caps. Free/trial product use is real, but it is not unlimited flagship API.
When free is enough vs when you need more: How to use Kimi K3 for free.

Source: Official Kimi Moon Landing Plan prize assets on the viral-referral share page.
Door B — You tried to pay and hit a wall
After launch, demand jammed Moonshot’s GPUs. On July 19, 2026, @Kimi_Moonshot said they were temporarily pausing new subscriptions and prioritizing compute for current members, while adding capacity and reopening spots in batches. They also signaled a future split between broader Kimi Membership and Kimi Code Membership.
What that means for you:
- Already subscribed → official line is you’re not the target of the pause.
- Trying to buy new → checkout may stay closed for a while; wait for batches.
- “I can’t subscribe” ≠ “K3 is dead.” Free product limits, existing members, and the separate paid API are different doors.
Full walkthrough: Kimi K3 subscription paused.
Door C — Call the API (kimi-k3)
Use the API when you need keys, tools, and a bill you can control—not when you only want a weekend chat.
Facts to wire (as of late July 2026 public docs):
| Item | Value |
|---|---|
| Model id | kimi-k3 |
| Context | 1M tokens (about a whole small project in one window—still bill carefully) |
| List price (USD / 1M tokens) | Cache hit $0.30 · input $3.00 · output $15.00 |
| Billing style | Flat pay-as-you-go; no context-length tiering on K3 |
| Platform | platform.kimi.ai · K3 pricing · K3 quickstart |
Rough first call shape (OpenAI-compatible; store keys in env, never in git):
export MOONSHOT_API_KEY="YOUR_KEY"
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["MOONSHOT_API_KEY"],
base_url="https://api.moonshot.ai/v1",
)
resp = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Outline a migration plan for a 200-file TypeScript monorepo."}],
)
print(resp.choices[0].message.content)
When K3 is worth the premium
- You keep hitting ~256K walls on older Kimi models
- Multi-hour agents, fat repos, vision-in-the-loop UI work
- You need production stability more than a free product quota
When to stay on cheaper coding models
- Tight IDE loops, PR-sized edits, high request volume
- Official overview still points day-to-day coding traffic at K2.7 Code / highspeed variants when speed and cost matter
Pricing walkthrough and switch rules: K3 release guide · K3 vs Claude & GPT.
Also note: platform docs flag older lines such as kimi-k2.5 and moonshot-v1 as closing to new users with a full sunset path toward August 31, 2026. If your stack still points there, plan a move—but K2.7 Code is not the same notice. Re-check model list before cutover.
Door D — Kimi Code CLI / coding agents
If you live in a terminal or coding agent:
- Start from the product/docs path on /kimi-code and Moonshot’s Kimi Code materials—not a random third-party “installer.”
- For routine shipping, keep
kimi-k2.7-code(or the highspeed sibling) as the default until your own evals say otherwise. - Bring
kimi-k3in for design-the-system / long-horizon / vision-feedback jobs—not every autocomplete. - Membership capacity for Code may diverge from general chat later (Moonshot’s batch-capacity note). Don’t assume one plan covers every surface forever.
Still waiting on July 27 open weights?
Moonshot’s public launch line: full model weights by July 27, 2026. As of July 24, treat that as a calendar promise, not a folder already on your disk.
Until a real Moonshot checkpoint shows up with license + README:
- You cannot honestly
ollama pullfull K3 - Random Hugging Face mirrors are noise until the official org posts files
- Even after weights land, 2.8T-class MoE is datacenter-shaped—not a single gaming GPU fantasy
Prep list: open weights July 27 · VRAM / Ollama reality.

Source: Official @Kimi_Moonshot K3 architecture media, July 16, 2026; also described on the Kimi K3 blog.
Quick decision tree (this week)
| If you are… | Do this |
|---|---|
| Curious human, first hour | Door A — free product + invite rewards |
| Blocked at paid checkout | Door B + free path; optional API if you ship code |
| Building a product/tool | Door C — budget kimi-k3; set spend caps |
| Shipping PRs all day | Door D — K2.7 Code first, K3 for hard jobs |
| Local-LLM hobbyist | Wait for verified weights; use API/product until then |
| Comparing “is K3 better than Claude?” | Pilot on your tasks; read vs Claude & GPT |
Watch-outs (common feed traps)
- “Open” ≠ free laptop model today. Product + API are live; full public weights are still on the July 27 promise.
- “Free API forever” is usually false. Product free tiers and paid
kimi-k3tokens are different products. - Subscription pause is capacity, not a scam. Confirm against the official X post and your own checkout UI.
- One benchmark crown is not a switch-all mandate. Frontend Arena strength or a Nature/Fortune headline does not mean you delete Claude keys tomorrow.
- This site is not Moonshot. We link primary sources; we don’t sell official keys or host weights.
Where next
- Specs & switch hub → /kimi-k3
- Living timeline → /kimi-k3-status
- Free vs paid myths → /blog/33-kimi-k3-free-how-to
- Pause details → /blog/35-kimi-k3-subscription-paused-access
- In-family pick → /blog/29-kimi-which-model-k26-k27-k3
- Kimi Code → /kimi-code
Want to try Kimi now? Create a free account and unlock rewards. Prefer membership when seats reopen? Subscribe for campaign perks.
Bottom line for the week of July 24: use the app/free door to learn, the API door to ship, keep K2.7 Code for everyday coding loops, and treat July 27 weights as a verify-then-celebrate event—not a reason to buy hardware on a rumor screenshot.