basetenlabs/K2-Vendor-Verifier

basetenlabs/K2-Vendor-Verifier

Description: Verify Precision of all Kimi K2 API Vendor

Stars: 0

Forks: 0

Open issues: 0

Created: 2025-11-17T20:33:43Z

Pushed: 2025-11-15T11:48:08Z

Default branch: main

Fork: yes

Parent repository: MoonshotAI/K2-Vendor-Verifier

Archived: no

README:

K2 Vendor Verifier

What's K2VV

Since the release of the Kimi K2 model, we have received numerous feedback on the precision of Kimi K2 in toolcall. Given that K2 focuses on the agentic loop, the reliability of toolcall is of utmost importance.

We have observed significant differences in the toolcall performance of various open-source solutions and vendors. When selecting a provider, users often prioritize lower latency and cost, but may inadvertently overlook more subtle yet critical differences in model accuracy.

These inconsistencies not only affect user experience but also impact K2's performance in various benchmarking results. To mitigate these problems, we launch K2 Vendor Verifier to monitor and enhance the quality of all K2 APIs.

We hope K2VV can help ensuring that everyone can access a consistent and high-performing Kimi K2 model.

K2-thinking Evaluation Results

Test Time: 2025-11-15

• temperature=1.0
• max_tokens=64000

Model Name provider ToolCall-Trigger Similarity ToolCall-Schema Accuracy

count_finish_reason_tool_calls count_successful_tool_call schema_accuracy

kimi-k2-thinking MoonshotAI - 1958 1958 100.00%

Moonshot AI Turbo >=73% 1984 1984 100.00%

InfiniAI 1827 1825 99.89%

SiliconFlow 2119 2097 98.96%

GMICloud 1850 1775 95.95%

AtlasCloud 1878 1798 95.74%

SGLang 1874 1790 95.52%

vLLM 2128 1856 87.22%

Parasail 2108 1837 87.14%

DeepInfra 2071 1800 86.91%

GoogleVertex 1945 1668 85.76%

Together 1893 1602 84.63%

NovitaAI 72.22% 1778 1715 96.46%

Chutes 68.10% 3657 3037 83.05%

Fireworks 67.38% 1494 1494 100.00%

##### We ran the official API multiple times to test the fluctuation of tool_call_f1. The lowest score was 75.81%, and the average was 76%. Given the inherent randomness of the model, we believe that an tool_call_f1 score above 73% is acceptable and can be used as a reference.

K2 0905 Evaluation Results

Test Time: 2025-11-15

• temperature=0.6

Model Name provider ToolCall-Trigger Similarity ToolCall-Schema Accuracy

count_finish_reason_tool_calls count_successful_tool_call schema_accuracy

kimi-k2-0905-preview MoonshotAI - 1274 1274 100.00%

Moonshot AI Turbo >=80% 1398 1398 100.00%

DeepInfra 1365 1365 100.00%

Fireworks 1453 1453 100.00%

Infinigence 1257 1257 100.00%

NovitaAI 1299 1299 100.00%

SiliconFlow 1305 1302 99.77%

Chutes 1271 1229 96.70%

vLLM 1325 1007 76.00%

SGLang 1269 928 73.13%

Volc 1330 969 72.86%

Baseten 1243 901 72.49%

AtlasCloud 1277 925 72.44%

Together 1266 911 71.96%

Groq 69.52% 1042 1042 100.00%

Nebius 50.60% 644 544 84.47%

##### We ran the official API multiple times to test the fluctuation of tool_call_f1. The lowest score was 82.71%, and the average was 84%. Given the inherent randomness of the model, we believe that an tool_call_f1 score above 80% is acceptable and can be used as a reference.

Evaluation Metrics

ToolCall-Trigger Similarity

We use tool_call_f1 to determine whether the model deployment is correct.

| Label / Metric | Formula | Meaning | | --- | --- | --- | | TP (True Positive) | — | Both model & official have finish_reason == "tool_calls". | | FP (False Positive) | — | Model finish_reason == "tool_calls" while official is "stop" or "others". | | FN (False Negative) | — | Model finish_reason == "stop" or "others" while official is "tool_calls". | | TN (True Negative) | — | Both model & official have finish_reason == "stop" or "others". | | tool_call_precision | TP / (TP + FP) | Proportion of triggered tool calls that should have been triggered. | | tool_call_recall | TP / (TP + FN) | Proportion of tool calls that should have been triggered and were. | | `tool_call_f1` | **2*tool_call_precision*tool_call_recall / (tool_call_precision+tool_call_recall) | Harmonic mean of precision and recall (primary metric for deployment check).** |

ToolCall-Schema Accuracy

We use schema_accuracy to measure the robustness of the engineering.

| Label / Metric | Formula / Condition | Description | | --- | --- | --- | | count_finish_reason_tool_calls | — | Number of responses with finish_reason == "tool_calls". | | count_successful_tool_call | — | Number of tool_calls responses that passed schema validation. | | `schema_accuracy` | `count_successful_tool_call / count_finish_reason_tool_calls` | Proportion of triggered tool calls whose JSON payload satisfies the schema. |

How we do the test

We test toolcall's response over a set of 4,000 requests. Each provider's responses are collected and compared against the official Moonshot AI API.

K2 vendors are periodically evaluated. If you are not on the list and would like to be included, feel free to contact us.

Sample Data: Detailed samples and MoonshotAI results are available in evaluation-dataset/tool-calls/ (50% of the test set).

Suggestions to Vendors

1. Use the Correct Versions Some vendors may not meet the requirements due to using incorrect versions. We recommend using the following versions and newer versions:

• K2-0905:
• vllm v0.11.0
• sglang v0.5.3rc0
• moonshotai/Kimi-K2-Instruct-0905 (commit: 94a4053eb8863059dd8afc00937f054e1365abbd)
• K2-thinking:
• vllm v0.11.1rc6
• sglang v0.5.5.post2
• moonshotai/Kimi-K2-Thinking latest

2. Rename Tool Call IDs The Kimi-K2 model expects all tool call IDs in historical messages to follow the format functions.func_name:idx. However, previous test cases may contain malformed tool IDs like serach:0*, which could mislead Kimi-K2 into generating incorrect tool call IDs, resulting in parsing failures. In this version, we manually add the functions. prefix to all previous tool calls to make Kimi-K2 happy :). We recommend that users and vendors adopt this fix in…