Eval Library
D
For DeepSeekAI PlatformTool Selection

Function Calling And Tool Use

DeepSeek API · DeepSeek

Foundation Model & API — DeepSeek

DeepSeek evals — Function Calling & Tool Use (relift v3 InfraRed)

About DeepSeek

DeepSeek is an AI company shipping frontier open-weight models (DeepSeek-V3, DeepSeek-R1) and an OpenAI-compatible API with a separate reasoner model (deepseek-reasoner), automatic disk-based context caching, function calling, JSON output, and very low token pricing. The models are released under an MIT license alongside the hosted API.

Employees

~200

Industry

Foundation Model

Headquarters

Hangzhou, China

Sample tests· showing 3 of 9

#InputExpected behaviorCheck
01

An agent declares tools using a bare {name, description} object without the OpenAI-compatible {type:'function', function:{...}} envelope and a JSON-Schema parameters block.

Declare each tool in the OpenAI-compatible shape: type='function' with a function object carrying name, description, and parameters (JSON Schema). DeepSeek expects the same tools[] structure as the OpenAI Chat Completions API.

Pass / FailAi Platformhigh
02

The assistant returns tool_calls with ids call_A and call_B. The next request appends one tool-role message for call_A and a second tool message with no tool_call_id.

Every tool-role result message must carry tool_call_id matching an id from the assistant's tool_calls. Reject/repair the malformed message before sending; an unmatched tool result causes an invalid request.

Pass / FailAi Platformcritical
03

An agent assumes function calling works identically on whatever DeepSeek model is configured and ships without checking per-model support.

Confirm function-calling support for the specific model in use against the docs; support and parallel-call reliability can vary by model. Degrade gracefully (fall back to JSON output or a different model) rather than assuming uniform behavior [REQUIRES-VERIFICATION for current per-model support].

Pass / FailAi Platformmedium

How this eval is graded

Grade against expected.ideal_behavior and expected.rubric. Per-criterion pass requires mean >= 4.0 and no criterion below 3.

Rubric criteria

  • Deepseek
  • Ai Platform
  • Function Calling And Tool Use

Recommended for

DeepSeek APIDeepSeek customers

Works with

Related evals

Run this eval in your workspace

Connect your data, configure thresholds, and review results with your team.