July 16, 2026·4 min read·Lee Penkman

Kimi K3: the largest open model, live on the app.nz gateway

Moonshot's 2.8T-parameter Kimi K3 with a flat-priced 1M context window is available now as kimi-k3 — direct Moonshot routing, automatic failover, one metered key.

Moonshot AI's Kimi K3 launched this week and it is available on the app.nz gateway today as kimi-k3. The headline specs: 2.8 trillion total parameters (the largest open-weights release to date), a 1,048,576-token context window with no length tiering, vision input, tool calling, and deep reasoning. Pricing is $3.00/M input and $15.00/M output, with $0.30/M on cache hits via Moonshot direct.

Early results back the size up: K3 debuted at #1 on the Frontend Code Arena ahead of Claude and GPT, and launch coverage positions it as the first open model genuinely closing on Opus 4.8.

Use it now

One metered key, OpenAI-compatible, no separate Moonshot account needed:

curl https://app.nz/api/gateway/v1/chat/completions \
  -H "Authorization: Bearer $APP_NZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3", "messages": [{"role": "user", "content": "Summarize this monorepo..."}]}'

from openai import OpenAI

client = OpenAI(base_url="https://app.nz/api/gateway/v1", api_key=APP_NZ_API_KEY)
resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Plan the migration in 5 steps"}],
    max_tokens=30000,
)

The gateway routes kimi-k3 to Moonshot's API directly and fails over automatically to OpenRouter (or/kimi-k3) and then Kimi K2.5, so a single upstream incident does not take your agents down. Usage is metered per token on your app.nz account like every other model in the catalog.

Where K3 fits

The 1M untiered window is the interesting part for agent workloads. Long-horizon coding agents burn most of their spend re-establishing context; a flat-priced 1M window plus $0.30 cached input changes the economics of keeping a whole repo in context across turns. K3's tool calling held up well in our gateway smoke tests, and reasoning currently runs at max effort by default (Moonshot says more levels are coming), so budget max_tokens generously for structured outputs — our pelican-on-a-bicycle SVG test blew through an 8K cap mid-path, then completed the full scene in ~7K tokens once given a 30K budget:

Kimi K3 one-shot SVG: pelican warrior riding a bicycle

Notes for agent builders

Alias kimi now points at K3 (K2.5 remains available as kimi-k2.5).
Vision input works through the standard OpenAI image content parts.
Open weights mean you can graduate to self-hosting later without prompt rework — same model, your GPUs.

K3 is live for every app.nz account now, prepay credits apply, and it is already selectable in coding-agent integrations that ride the gateway.

Use it now

One metered key, OpenAI-compatible, no separate Moonshot account needed:

curl https://app.nz/api/gateway/v1/chat/completions \
  -H "Authorization: Bearer $APP_NZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3", "messages": [{"role": "user", "content": "Summarize this monorepo..."}]}'

from openai import OpenAI

client = OpenAI(base_url="https://app.nz/api/gateway/v1", api_key=APP_NZ_API_KEY)
resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Plan the migration in 5 steps"}],
    max_tokens=30000,
)

Where K3 fits

Kimi K3 one-shot SVG: pelican warrior riding a bicycle

Notes for agent builders

Alias kimi now points at K3 (K2.5 remains available as kimi-k2.5).

Vision input works through the standard OpenAI image content parts.

Open weights mean you can graduate to self-hosting later without prompt rework — same model, your GPUs.

K3 is live for every app.nz account now, prepay credits apply, and it is already selectable in coding-agent integrations that ride the gateway.