On this page
01
Use cases
What this model is useful for
Compare documentation and code
Label excerpts with filenames or section IDs. Require the exact passage supporting or contradicting each conclusion.
Find a missing test
Provide the contract, handler and existing tests. Keep a proposed test separate from one that has actually run.
Review a long document
Start with concrete questions and required sections. Omissions are easier to spot than in a generic summary of dozens of pages.
02
Examples
Prompts you can try
A sample brief — add your own inputs. This page does not generate content or charge your balance.
Prompt for an independent test — no saved Kimi K3 response or executed retry is attached.
An idempotency contract
All three excerpts are synthetic. In a real project, use redacted material and exclude keys, tokens and personal data.
Given these excerpts:
[DOC-1] POST /jobs accepts Idempotency-Key; a retry with the same key returns the original job_id.
[CODE-1] the handler always calls create_job() and never reads headers.
[TEST-1] only one request is covered.
Return a table with claim, supporting excerpt, contradiction and missing test. Do not claim that a retry was actually executed.Make the prompt your own
- Provide the reviews without your ratings or expected conclusions to avoid steering the answer.
- If you need counts, define the counting rule and check the totals.
How to check the result
- The output references DOC-1, CODE-1 and TEST-1 without inventing other excerpts.
- The contradiction between promised idempotency and unconditional create_job() is stated directly.
- The missing test uses two requests with the same key but is not presented as executed.
- A published run preserves the full input, output, model ID, date, test evidence and usage.
03
Pricing / USD
CLODEx public pricing
The card shows CLODEx pricing for kimi-k3. Open weights do not make a hosted API call free; a CLODEx route and local deployment have different costs.
Full pricing details04
Get started
How to get started
Check the ID
Confirm that kimi-k3 appears in /v1/models for your key. The public page does not prove availability for every account.
Configure a compatible client
Use the CLODEx base URL and authentication from /docs, then select the exact kimi-k3 ID.
Start with three excerpts
Test the short labeled fixture first. Then add real files in stages while preserving source references.
One API for your application.
curl --fail-with-body --no-buffer 'https://clodex.xyz/v1/chat/completions' \
-H "Authorization: Bearer $CLODEX_API_KEY" \
-H 'Content-Type: application/json' \
--data-binary @- <<'CLODEX_JSON'
{"messages":[{"content":"Given these excerpts:\n[DOC-1] POST /jobs accepts Idempotency-Key; a retry with the same key returns the original job_id.\n[CODE-1] the handler always calls create_job() and never reads headers.\n[TEST-1] only one request is covered.\n\nReturn a table with claim, supporting excerpt, contradiction and missing test. Do not claim that a retry was actually executed.","role":"user"}],"model":"kimi-k3","stream":true}
CLODEX_JSONKeep your key on the server, never in the browser. A generation example uses your balance only when you run it yourself. Check model access and parameters before submitting.
How to get an API key05
Specifications
Capabilities and limitations
- Model ID
kimi-k3- API protocols
- OpenAI-compatible API
Only parameters declared by the integration catalog are shown. An omitted parameter is not a promise of support.
Model and integration
What this model is
Moonshot AI publishes Kimi K3 as an open-weights model for code, documents and tools. CLODEx exposes a hosted route under the kimi-k3 ID.
Why excerpt labels matter
Labels such as DOC-1, CODE-1 and TEST-1 make every conclusion traceable. Without them, a long confident answer is harder to distinguish from an incomplete summary.
How to handle large material
Provide sections relevant to the goal rather than an entire database. Remove secrets and personal data, ask specific questions and preserve unknowns.
What to consider
Context does not guarantee completeness
Even if a document fits in one request, the model may miss an exception or confuse sections. Use a checklist of required points.
Open weights and API are different
Moonshot's card describes the model, while this page covers a hosted CLODEx route. Local deployment, hardware and API pricing are different concerns.
Tools and images need verification
Model capability does not mean the CLODEx endpoint and client accept the same format. Verify each feature separately.
Before you submit
A large context window does not guarantee every condition will be found. Provide only relevant redacted excerpts, label them and verify every source reference.
06
Related models
Models to compare
GLM-5.3
glm-5.3Use the same excerpts and checklist for code and documents. Compare citation accuracy, omissions and cost per accepted answer.
Claude Fable 5.1
claude-fable-5.1Useful for comparing long document work. Keep the file set and evaluation criteria identical across runs.
07
FAQ
Model FAQ
How is API access different from local deployment?
CLODEx provides a hosted route billed at its rate. Running open weights locally requires your own hardware and operations.
Do I need a Kimi subscription?
No. A Kimi app subscription and API access through CLODEx are different products. This page uses a CLODEx key and balance.
Can I send a very large document?
Check the route limit, but do not rely on context size alone. Split the document, mark required sections and verify references.
Are images supported?
The upstream card describes image input, but the selected CLODEx route must support the format.
Where do I get an API key?
Create a key on the CLODEx API Keys page and follow /docs. If you connect from Russia, separately check current service availability, payment methods and data-processing terms.
Can I trust a reported contradiction?
Only after checking the cited excerpts and running the missing test. The model can formulate a hypothesis but does not replace system verification.
08
