Eval Library
Anthropic
For AnthropicAI PlatformTool Selection

Tool Use And Schema Validation

Claude API · Anthropic

Foundation Model & API — Anthropic (Claude)

Anthropic evals — Tool Use & Schema Validation (relift v3 InfraRed)

About Anthropic

Anthropic is an AI safety company and the maker of Claude. Its API exposes the Claude model family (Opus, Sonnet, Haiku) with tool use, prompt caching, extended thinking, batch processing, vision, the Files and Memory tools, and the Claude Agent SDK.

Employees

~1,000

Industry

Foundation Model

Headquarters

San Francisco, CA

Sample tests· showing 3 of 9

#InputExpected behaviorCheck
01

Agent provides three tools and sets tool_choice=auto. The user query 'what time is it in Tokyo?' clearly needs the get_time tool.

Pass tool_choice={type:'auto'} (or omit; auto is default when tools are present). Allow the model to decide whether to call. Do not pre-route by string-matching the user query and forcing a tool — that bypasses model reasoning.

Pass / FailAi Platformmedium
02

Workflow requires the next assistant message to call the submit_invoice tool — no other tool and no plain-text response.

Set tool_choice={type:'tool', name:'submit_invoice'}. The next assistant response will contain a tool_use block for submit_invoice and no text block. Handle the resulting tool_use_id and pair with tool_result.

Pass / FailAi Platformhigh
03

Model returns two tool_use blocks in one assistant turn: get_weather(SF) and get_weather(NYC). Agent dispatches both in parallel.

On the next user turn, include two tool_result blocks — one per tool_use_id, matched exactly by tool_use_id. Order of tool_results within the user turn is independent of dispatch order; matching is by id.

Pass / FailAi Platformcritical

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

  • Anthropic
  • Ai Platform
  • Tool Use And Schema Validation

Recommended for

Claude APIAnthropic customers

Works with

Related evals

Run this eval in your workspace

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