feat(llm): add moonshot/kimi-k3 to model prices and context window map

[tin-berri]

TLDR

Problem this solves:

• moonshot/kimi-k3 is missing from the model price map
• Requests to it fail model-info lookup and track $0 spend

How it solves it:

• Adds moonshot/kimi-k3 to model_prices_and_context_window.json (+ backup copy)
• Official pricing: $3.00/M input, $0.30/M cache read, $15.00/M output, 1,048,576 context

User Flow

Before: a developer routing Kimi K3 through the gateway gets no pricing metadata, so spend tracking reads zero

1. They send POST https://litellm-domain/v1/chat/completions with "model": "moonshot/kimi-k3"
2. The response comes back fine, but the spend logs show $0.00 for the request
3. Calling GET https://litellm-domain/v1/model/info for the model shows no max_input_tokens or cost fields
4. https://litellm-domain/ui/?page=logs shows the request at $0.00 spend

After: the same request is priced correctly

1. They send POST https://litellm-domain/v1/chat/completions with "model": "moonshot/kimi-k3"
2. The response comes back, and the spend logs show real cost ($3.00/M input, $15.00/M output, $0.30/M cache hits)
3. GET https://litellm-domain/v1/model/info returns 1,048,576 max input/output tokens plus reasoning/vision/tool-calling support flags
4. https://litellm-domain/ui/?page=logs shows the request at non-zero spend

Relevant issues

No issue was filed for the moonshot entry: kimi-k3 was noticed absent while configuring it as a gateway model. Related: #35344 asks for the Fireworks kimi-k3 entry, which is a separate change

Linear ticket

Resolves LIT-5873

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests (n/a: data-only change to the price map)
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more (for this data-only change, the same checks CI runs pass locally: jq empty, ci_cd/check_files_match.py, ci_cd/generate_model_prices_schema.py --check)
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Full before/after QA, live against Moonshot: the before leg at the merge base (9821b45) prices a real completion at $0, the after leg at the PR tip (277ed88) prices the same completion from the new entry. Each leg ran on its own proxy booted from that commit with LITELLM_LOCAL_MODEL_COST_MAP=True, its own Postgres, and a virtual key

Pricing source: https://platform.kimi.ai/docs/pricing/chat-k3 ($3.00/M cache-miss input, $0.30/M cache-hit input, $15.00/M output, 1,048,576-token context). Capabilities per https://platform.kimi.ai/docs/guide/kimi-k3-quickstart.md: max_completion_tokens settable up to 1,048,576, native vision and video file input, always-on reasoning with reasoning_effort (low/high/max)

Before, at the merge base (9821b45)

1. Live chat completion on a proxy booted from this commit

curl -sS -w '\nHTTP_STATUS=%{http_code}\n' http://localhost:57773/v1/chat/completions \
  -H 'Authorization: Bearer sk-qa-37552-before' -H 'Content-Type: application/json' \
  -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Reply with exactly: ok"}],"max_tokens":10}'

HTTP_STATUS=200, a real completion (id chatcmpl-6a88dda82e203c136a972943, 90 prompt + 10 completion tokens)

2. The spend surface logs that request at $0

curl -sS -w '\nHTTP_STATUS=%{http_code}\n' \
  'http://localhost:57773/spend/logs?request_id=chatcmpl-6a88dda82e203c136a972943' \
  -H 'Authorization: Bearer sk-master-qa37552-before'

HTTP_STATUS=200, one row (relevant fields):

"model": "moonshot/kimi-k3",
"custom_llm_provider": "moonshot",
"spend": 0.0,
"prompt_tokens": 90,
"completion_tokens": 10,
"total_tokens": 100

3. GET /v1/model/info on the same proxy: HTTP_STATUS=200 with model_info for kimi-k3 all null (relevant fields):

"key": "moonshot/kimi-k3",
"litellm_provider": "moonshot",
"max_input_tokens": null,
"max_output_tokens": null,
"max_tokens": null,
"mode": null,
"input_cost_per_token": 0,
"output_cost_per_token": 0,
"cache_read_input_token_cost": null,
"supports_vision": null,
"supports_function_calling": null

Every other cost field and supports_* flag also came back null

4. SDK at the same commit

LITELLM_LOCAL_MODEL_COST_MAP=True python -c '
import litellm
litellm.get_model_info("moonshot/kimi-k3")
litellm.cost_per_token(model="moonshot/kimi-k3", prompt_tokens=1000000, completion_tokens=100000)'

get_model_info -> Exception: This model isn't mapped yet. model=moonshot/kimi-k3, custom_llm_provider=moonshot. Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json.
cost_per_token -> Exception: This model isn't mapped yet. model=moonshot/kimi-k3, custom_llm_provider=moonshot. Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json.

After, at the PR tip (277ed88)

1. The same live chat completion on a proxy booted from this commit

curl -sS -w '\nHTTP_STATUS=%{http_code}\n' http://localhost:48290/v1/chat/completions \
  -H 'Authorization: Bearer sk-qa-37552-after' -H 'Content-Type: application/json' \
  -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Reply with exactly: ok"}],"max_tokens":10}'

HTTP_STATUS=200, a real completion (id chatcmpl-6a88db3f50c94b5b8f5a3029, 90 prompt + 10 completion tokens), and the response now carries cost headers

x-litellm-response-cost: 0.00042
x-litellm-response-cost-input: 0.00027
x-litellm-response-cost-output: 0.00015

2. The spend surface logs real cost priced from the new entry

curl -sS -w '\nHTTP_STATUS=%{http_code}\n' \
  'http://localhost:48290/spend/logs?request_id=chatcmpl-6a88db3f50c94b5b8f5a3029' \
  -H 'Authorization: Bearer sk-master-qa37552-after'

HTTP_STATUS=200, one row (relevant fields):

"model": "moonshot/kimi-k3",
"custom_llm_provider": "moonshot",
"spend": 0.00042,
"prompt_tokens": 90,
"completion_tokens": 10,
"total_tokens": 100

90 * $3.00/M input + 10 * $15.00/M output = $0.00042, matching the logged spend exactly, and the row's model_map_information names moonshot/kimi-k3 with input_cost_per_token: 3e-06

3. GET /v1/model/info on the same proxy: HTTP_STATUS=200, model_info for kimi-k3 (relevant fields verbatim):

"key": "moonshot/kimi-k3",
"litellm_provider": "moonshot",
"max_input_tokens": 1048576,
"max_output_tokens": 1048576,
"max_tokens": 1048576,
"mode": "chat",
"input_cost_per_token": 3e-06,
"output_cost_per_token": 1.5e-05,
"cache_read_input_token_cost": 3e-07,
"supports_reasoning": true,
"supports_vision": true,
"supports_function_calling": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_video_input": true

4. SDK at the same commit

LITELLM_LOCAL_MODEL_COST_MAP=True python -c '
import litellm, json
print(json.dumps(litellm.get_model_info("moonshot/kimi-k3"), indent=1))
p, c = litellm.cost_per_token(model="moonshot/kimi-k3", prompt_tokens=1000000, completion_tokens=100000)
print(f"prompt cost: ${p:.2f}, completion cost: ${c:.2f}, total: ${p+c:.2f}")'

{
 "max_input_tokens": 1048576,
 "max_output_tokens": 1048576,
 "input_cost_per_token": 3e-06,
 "cache_read_input_token_cost": 3e-07,
 "output_cost_per_token": 1.5e-05,
 "litellm_provider": "moonshot",
 "mode": "chat",
 "supports_vision": true,
 "supports_reasoning": true,
 "supports_function_calling": true
}
prompt cost: $3.00, completion cost: $1.50, total: $4.50

Observations from the run, in scope for a reviewer but not caused by this PR:

• supports_prompt_caching stays null; matches the JSON entry, left alone
• Pre-existing /v1/messages moonshot prefix bug ([Bug]: /v1/messages drops the vendor half of perplexity/perplexity/* model ids — Agent API models 400, sonar silently answers off the chat endpoint #37716), untouched here
• Cache-read pricing not exercised live; entry matches docs
• kimi-k3 reasoning consumed all 10 max_tokens; content empty
• /spend/logs request_id filter wants the response id (both legs)

Type

🆕 New Feature

Caveats (if any)

• supports_web_search deliberately omitted: K3 docs say web_search is being updated (matches kimi-k2.5/k2.6)

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

---

Note

Low Risk
Data-only price-map entry; no runtime, auth, or routing logic changes. Wrong numbers would only affect cost reporting.

Overview
Adds moonshot/kimi-k3 to the model price map (and backup) so LiteLLM can resolve costs and context limits instead of treating the model as unmapped ($0 spend).

Pricing is $3.00/M input, $0.30/M cache read, $15.00/M output, with a 1,048,576-token window. Capability flags match other recent Kimi chat models (tools, reasoning, vision/video, response schema).

^{Reviewed by Cursor Bugbot for commit 277ed88. Bugbot is set up for automated code reviews on this repo. Configure here.}

• 277ed88 passes /live-pr-risk

[greptile-apps]

Greptile Summary

Adds Moonshot Kimi K3 pricing, context-window, and capability metadata to the canonical model map and its backup copy.

• Configures input, cache-read, and output token costs.
• Declares a 1,048,576-token context window.
• Keeps the canonical and backup entries synchronized.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds the canonical Moonshot Kimi K3 pricing and model-capability entry using existing schema fields.
litellm/model_prices_and_context_window_backup.json	Mirrors the canonical Kimi K3 metadata entry in the backup price map.

_{Reviews (3): Last reviewed commit: "Merge remote-tracking branch 'origin/lit..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_add_kimi_k3_pricing (277ed88) with litellm_internal_staging (04113aa)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (9821b45) during the generation of this report, so 04113aa was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 42cffe9. Configure here.}

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 277ed88. Configure here.}

[mateo-berri]

LGTM. Thanks!