Build fluency with advanced Claude API concepts — batching, tool use content blocks, extended thinking, and streaming patterns.
0 / 5 completed
1 / 5
At standup, a colleague asks what the Messages Batching API is designed for. What is the correct answer?
The Messages Batching API is designed for asynchronous batch processing: you submit up to 10,000 requests in a single call and either poll for completion or retrieve results when the batch finishes. It trades latency for throughput and cost efficiency, and does not bypass rate limits or enable real-time streaming.
2 / 5
During a PR review, a teammate asks what a tool_use content block in an assistant response represents. What is the correct description?
A tool_use content block in the assistant response contains the name of the tool, the input JSON, and a tool_use_id. The caller must execute the tool and return the result in a tool_result content block that echoes back the same tool_use_id so the model can correlate the result.
3 / 5
In a design review, the team discusses extended thinking in Claude. Which statement correctly describes this feature?
Extended thinking allows Claude to reason internally before producing its final answer. It is enabled by passing thinking: { type: "enabled", budget_tokens: N } in the request. The response includes thinking blocks containing the internal reasoning, followed by the final text block. It is not a caching or fine-tuning mechanism.
4 / 5
An incident report shows a request failing with a 400 error referencing a missing beta header. What is the correct explanation?
Some Claude capabilities — including extended thinking and computer use — are gated behind an anthropic-beta HTTP header. You must include the appropriate beta feature flag (e.g. interleaved-thinking-2025-05-14) in the header, otherwise the API returns a 400 error indicating the feature is not available without the opt-in header.
5 / 5
During a code review, a senior engineer asks about streaming with tool calls and specifically what input_json_delta events represent. What is correct?
When streaming a response that contains a tool call, Claude emits input_json_delta events carrying incremental chunks of the tool's JSON input. Callers must concatenate these deltas to reconstruct the complete input object before executing the tool. This mirrors how text_delta events work for text content.