Kimi K3.1 Pro on Wallaby: 10% off and agent setup

Kimi K3.1 Pro is live on Wallaby at 10% off the official list.

The model ID is kimi-k3.1-pro, and it is enabled for the default user group now. Standard pricing is $0.18 cached input, $1.80 cache-miss input and $9.00 output per 1M tokens.

You can call it through the same OpenAI-compatible endpoint used by our other guides, https://api.wallabytoken.com/v1. The current public rate card lists it as a standing model, and the request below is a direct way to verify your own key.

Disclosure: Wallaby provides inference-as-a-service for curated large language models. We checked production enablement, request logs, the npm registry and the models.dev catalog on 5 November 2026. This article separates what works today from what still needs a package update.

K3.1 Pro is a reasoning model with a 1,048,576-token context. It supports tool calling and streaming. Streaming responses separate reasoning_content deltas from final-answer content deltas, so an agent can display or log the two streams separately. reasoning_effort accepts low, high and max, with max as the default.

The price, in one table

Standard K3.1 Pro costs 90% of Moonshot's official list price for the same three billing items.

These are standing rates, not a launch countdown. The table covers the three confirmed billing items. Separate 5-minute and 1-hour cache-write pricing is not included because our current public rate card does not publish those two items.

K3.1 Pro Highspeed is a separate model ID with a different rate. Cached input is also $0.18, while cache-miss input and output are $3.60 and $18.00 per 1M tokens. This article is about the standard kimi-k3.1-pro model. For the full rate card and worked cost examples, see Kimi K3.1 Pro pricing.

Make your first K3.1 Pro call

New accounts receive $0.50 in free credit. Once that trial credit is nearly exhausted, an eligible first top-up is $5 once, after which the standard minimum is $10. The $5 tier requires an account at least 24 hours old and no prior successful top-up. Prepaid credit is not withdrawable.

Export your key and send a minimal request:

export WALLABY_API_KEY="sk-your-wallaby-key"

curl https://api.wallabytoken.com/v1/chat/completions \
  -H "Authorization: Bearer $WALLABY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3.1-pro",
    "messages": [
      {"role": "user", "content": "Reply with exactly: OK"}
    ]
  }'

The OpenAI Python SDK uses the same endpoint and model ID:

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-wallaby-key",
    base_url="https://api.wallabytoken.com/v1",
)

response = client.chat.completions.create(
    model="kimi-k3.1-pro",
    messages=[{"role": "user", "content": "Reply with exactly: OK"}],
)

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

After the call, open Usage Logs in the console. Each request is itemized with model, token counts and cost, so you can reconcile the client against the invoice instead of guessing from a status bar.

Agent support: what works today

Which setup is cleanest depends on your agent. Some have a provider entry or package; others accept the OpenAI-compatible endpoint directly. The important distinction is that the production model is live, while package metadata can lag behind it.

OpenAI-compatible clients use the /v1 prefix; Codex uses /v1/responses. Claude Code and Goose's OPENAI_HOST take the host without /v1.

A guide's verification date matters. A working K3 setup proves the connection path, but it does not prove that every client has been retested with K3.1 Pro. Where a package or catalog has not caught up, use cURL, the OpenAI SDK or the manual OpenAI-compatible configuration instead of waiting for the metadata update.

Choosing your path

Use Cline's OpenAI Compatible provider if you want a graphical setup. Use Codex CLI if you already work from a terminal and want a profile file. Use OpenHands if you prefer saved LLM profiles in a GUI. Use Claude Code if you need the Anthropic-compatible protocol rather than OpenAI chat completions.

For a scripted evaluation, start with cURL or the OpenAI SDK. That takes the client out of the equation: if the call works, K3.1 Pro is available and the remaining work is integration metadata.

For team rollouts, one account is enough. Give each developer a named key with its own cap and optional expiry, then use per-key logs to attribute spend. At the Wallaby gateway layer, we don't store your prompts, we don't train on your data, and we don't publish your usage history. Teams spending over $500 a month can contact [email protected] for volume pricing and priority support.

Two npm packages: current support

There are two Wallaby npm packages in the current integration line:

The package names stay the same. For pi, newly enabled models with a pricing.json entry can appear after a restart without another npm release. The npm registry remains the source of truth for shipped package versions.

Two adjacent updates are separate from npm:

• The models.dev catalog needs a K3.1 Pro entry before OpenCode can present it through the current catalog path.
• The Hermes and Letta integrations are provider plugins, not npm packages. Both read the gateway's /models endpoint, so they list K3.1 Pro without a package update; their fallback lists are only for failed probes.

That keeps the claim honest: the model is callable today; each client gets its own verified update.

FAQ

Is standard K3.1 Pro actually available to users now?

Yes. The production default group has kimi-k3.1-pro enabled. The standard rate is $0.18 cached input, $1.80 cache-miss input and $9.00 output per 1M tokens.

Does every agent support K3.1 Pro today?

No. The API is live, but catalog and package metadata can lag behind it. Codex CLI, OpenHands, Claude Code, pi 0.2.0, Goose, Hermes, Letta and wallaby-ai-provider 0.1.1 work today. OpenCode's models.dev catalog and Cline's built-in provider list still need updates.

What is the difference between K3.1 Pro and Highspeed?

They are different model IDs and rate cards. This article covers standard kimi-k3.1-pro; it does not imply that Highspeed shares the same rate.

Do the npm packages support K3.1 Pro?

[email protected] supports it now. [email protected] already passes the model ID through, so no package update is required for basic use.

What you have now, and where to go next

You now have the model ID, endpoint, confirmed rate and a working path for your agent. For a deeper breakdown, read the K3.1 Pro rate card. For other client setups, use the Kimi CLI and agent integration hub. To make a first call, create an account and get your key.

Sources

• Moonshot model inference pricing
• Wallaby public pricing.json
• pi-wallaby-provider on npm
• wallaby-ai-provider on npm
• models.dev API catalog