Back to evaluations
Public evaluation

ComplexQuerySynthesis

The evaluation will focus on the accuracy, completeness, and verifiability of the synthesized answers, as well as the efficiency of the hybrid reasoning process. It will assess how well the system uses its tools and adaptive budgets.

Evaluation type
task based
Challenge
Hybrid Reasoning AI Browser for Real-time Synthesis with Gemini 2.5 Pro & DSPy
Difficulty
Advanced
Rigor
Unspecified

Evaluation overview

How the linked challenge is judged: tasks, benchmarks, and criteria count.

Tasks
2
Benchmarks
0
Criteria
0

Task templates

Inputs and expected outputs.

Task 1

ComplexQuerySynthesis

Provide a complex, multi-faceted query requiring web browsing and synthesis from multiple sources.

Input format

String representing the user query.

Output format

JSON object with 'answer', 'sources' (list of URLs), 'reasoning_trace' (steps taken by agents).

Task 2

AdaptiveBudgetEfficiency

Submit queries of varying complexity and evaluate the token usage and response time.

Input format

String representing the user query.

Output format

JSON object with 'query', 'token_usage_total', 'response_time_ms', 'deep_reasoning_invoked'.