Natural Questions vs cua
Side-by-side comparison to help you choose.
| Feature | Natural Questions | cua |
|---|---|---|
| Type | Dataset | Agent |
| UnfragileRank | 48/100 | 53/100 |
| Adoption | 1 | 1 |
| Quality | 0 | 1 |
| Ecosystem | 0 |
| 1 |
| Match Graph | 0 | 0 |
| Pricing | Free | Free |
| Capabilities | 6 decomposed | 15 decomposed |
| Times Matched | 0 | 0 |
Evaluates end-to-end QA systems by requiring models to both retrieve relevant Wikipedia passages from 5.9M articles and extract answers from those passages. Unlike single-document QA benchmarks, Natural Questions forces systems to solve the full information retrieval pipeline before reading comprehension, using real Google Search queries as ground truth for relevance. Annotators provide both paragraph-level (long answer) and entity-level (short answer) labels, enabling fine-grained performance measurement across retrieval and extraction stages.
Unique: Combines retrieval and reading comprehension in a single benchmark using real Google Search queries, forcing systems to solve the full open-domain QA pipeline rather than isolated reading comprehension on pre-selected passages. The dual-annotation scheme (long + short answers) enables separate measurement of retrieval quality and extraction accuracy.
vs alternatives: More realistic than SQuAD (which provides passage context) because it requires actual retrieval; more comprehensive than MS MARCO (which focuses on ranking) because it evaluates end-to-end answer extraction from retrieved passages
Provides two complementary answer labels per question: long answers (full paragraph from Wikipedia containing the answer) and short answers (minimal entity or phrase). This dual-level annotation enables training and evaluating both passage-ranking and span-extraction components separately. Annotators mark questions as unanswerable if no Wikipedia article contains the answer, creating a realistic distribution of answerable vs. unanswerable queries matching production search logs.
Unique: Dual-level annotation (paragraph + entity) decouples retrieval evaluation from reading comprehension, allowing separate optimization of passage ranking and span extraction. The explicit unanswerable label distribution reflects real search query distributions rather than assuming all questions have answers.
vs alternatives: More granular than SQuAD's single-span annotation because it separates passage retrieval from answer extraction; more realistic than MS MARCO because it includes explicit unanswerable examples matching production query distributions
Dataset contains 307,373 real, anonymized queries extracted from Google Search logs, ensuring the question distribution reflects actual user information needs rather than synthetic or crowdsourced questions. This ground-truth distribution includes long-tail queries, ambiguous questions, and unanswerable searches that production systems must handle. Pairing these queries with Wikipedia articles creates a realistic open-domain QA evaluation setting where systems must handle the full diversity of real user intent.
Unique: Uses real Google Search queries rather than crowdsourced or synthetic questions, capturing the true distribution of user information needs including long-tail, ambiguous, and unanswerable searches. This grounds evaluation in production-grade query patterns rather than benchmark-specific biases.
vs alternatives: More representative of real user intent than SQuAD or MS MARCO because it derives from actual search logs; captures natural query diversity and ambiguity that synthetic benchmarks cannot replicate
Provides a fixed corpus of 5.9M Wikipedia articles as the knowledge base for retrieval evaluation. Systems must rank and retrieve relevant articles/passages from this corpus to answer questions, enabling measurement of retrieval quality (recall@k, MRR) independent of reading comprehension. The corpus is structured with article-level and paragraph-level granularity, allowing evaluation of both coarse document retrieval and fine-grained passage ranking. This setup forces realistic retrieval challenges: handling polysemy, disambiguation, and ranking relevant passages above irrelevant ones from the same article.
Unique: Provides a large, fixed Wikipedia corpus (5.9M articles) with paragraph-level granularity, enabling evaluation of both document-level and passage-level retrieval. The corpus size and diversity force systems to handle realistic retrieval challenges like disambiguation and ranking relevant passages above irrelevant ones from the same article.
vs alternatives: Larger and more diverse than MS MARCO's passage corpus because it covers all of Wikipedia; more realistic than SQuAD because it requires actual retrieval rather than providing context upfront
Explicitly labels ~20% of questions as unanswerable (no Wikipedia article contains the answer), enabling evaluation of systems' ability to recognize when they cannot answer a question rather than hallucinating. This answerability classification is crucial for production systems that must gracefully handle out-of-domain or factually impossible queries. The distribution of answerable vs. unanswerable questions reflects real search query patterns, not synthetic balanced datasets.
Unique: Explicitly includes unanswerable questions (~20%) with ground-truth labels, enabling direct evaluation of systems' ability to recognize when they cannot answer. This reflects real query distributions where many searches have no valid answer in any single knowledge base.
vs alternatives: More realistic than SQuAD or MS MARCO because it includes explicit unanswerable examples; forces systems to avoid hallucination rather than assuming all questions have answers
Enables training and evaluating modular QA systems with separate retrieval and reading comprehension stages. The dataset structure (questions paired with Wikipedia corpus and dual-level answer annotations) supports training a dense retriever on passage relevance, a reader on span extraction, and an answerability classifier on unanswerable queries. Evaluation can measure each stage independently (retrieval recall, reader F1, answerability accuracy) or end-to-end (final answer accuracy), enabling fine-grained performance analysis and bottleneck identification.
Unique: Dataset structure explicitly supports training and evaluating modular QA pipelines with separate retrieval and reading comprehension stages. Dual-level annotations (long + short answers) and answerability labels enable independent optimization and evaluation of each component.
vs alternatives: More suitable for modular pipeline training than end-to-end QA datasets because it provides both passage-level and answer-level labels; enables separate measurement of retrieval and comprehension unlike single-stage QA benchmarks
Captures desktop screenshots and feeds them to 100+ integrated vision-language models (Claude, GPT-4V, Gemini, local models via adapters) to reason about UI state and determine appropriate next actions. Uses a unified message format (Responses API) across heterogeneous model providers, enabling the agent to understand visual context and generate structured action commands without brittle selector-based logic.
Unique: Implements a unified Responses API message format abstraction layer that normalizes outputs from 100+ heterogeneous VLM providers (native computer-use models like Claude, composed models via grounding adapters, and local model adapters), eliminating provider-specific parsing logic and enabling seamless model swapping without agent code changes.
vs alternatives: Broader model coverage and provider flexibility than Anthropic's native computer-use API alone, with explicit support for local/open-source models and a standardized message format that decouples agent logic from model implementation details.
Provisions isolated execution environments across macOS (via Lume VMs), Linux (Docker), Windows (Windows Sandbox), and host OS, with unified provider abstraction. Handles VM/container lifecycle (creation, snapshot management, cleanup), resource allocation, and OS-specific action handlers (keyboard/mouse events, clipboard, file system access) through a pluggable provider architecture that abstracts platform differences.
Unique: Implements a pluggable provider architecture with unified Computer interface that abstracts OS-specific action handlers (macOS native events via Lume, Linux X11/Wayland via Docker, Windows input simulation via Windows Sandbox API), enabling single agent code to target multiple platforms. Includes Lume VM management with snapshot/restore capabilities for deterministic testing.
vs alternatives: More comprehensive OS coverage than single-platform solutions; Lume provider offers native macOS VM support with snapshot capabilities unavailable in Docker-only alternatives, while unified provider abstraction reduces code duplication vs. platform-specific agent implementations.
cua scores higher at 53/100 vs Natural Questions at 48/100. Natural Questions leads on adoption, while cua is stronger on quality and ecosystem.
Need something different?
Search the match graph →© 2026 Unfragile. Stronger through disorder.
Provides Lume provider for provisioning and managing macOS virtual machines with native support for snapshot creation, restoration, and cleanup. Handles VM lifecycle (boot, shutdown, resource allocation) with optimized startup times. Integrates with image registry for VM image management and caching. Supports both Apple Silicon and Intel Macs. Enables deterministic testing through snapshot-based environment reset between agent runs.
Unique: Implements Lume provider with native macOS VM management including snapshot/restore capabilities for deterministic testing, optimized startup times, and image registry integration. Supports both Apple Silicon and Intel Macs with unified provider interface.
vs alternatives: More efficient than Docker for macOS because Lume uses native virtualization (Virtualization Framework) vs. Docker's slower emulation; snapshot/restore enables faster environment reset vs. full VM recreation.
Provides command-line interface (CLI) for quick-start agent execution, configuration, and testing without writing code. Includes Gradio-based web UI for interactive agent control, real-time monitoring, and trajectory visualization. CLI supports task specification, model selection, environment configuration, and result export. Web UI enables non-technical users to run agents and view execution traces with HUD visualization.
Unique: Implements both CLI and Gradio web UI for agent execution, with CLI supporting quick-start scenarios and web UI enabling interactive control and real-time monitoring with HUD visualization. Reduces barrier to entry for non-technical users.
vs alternatives: More accessible than SDK-only frameworks because CLI and web UI enable non-developers to run agents; Gradio integration provides quick UI prototyping vs. custom web development.
Implements Docker provider for running agents in containerized Linux environments with full isolation. Handles container lifecycle (creation, cleanup), image management, and volume mounting for persistent storage. Supports custom Dockerfiles for environment customization. Provides X11/Wayland display server integration for GUI application interaction. Enables reproducible agent execution across different host systems.
Unique: Implements Docker provider with X11/Wayland display server integration for GUI application interaction, container lifecycle management, and custom Dockerfile support. Enables reproducible agent execution across different host systems with container isolation.
vs alternatives: More lightweight than VMs because Docker uses container isolation vs. full virtualization; X11 integration enables GUI application support vs. headless-only alternatives.
Implements Windows Sandbox provider for isolated agent execution on Windows 10/11 Pro/Enterprise, and host provider for direct OS execution. Windows Sandbox provider creates ephemeral sandboxed environments with automatic cleanup. Host provider enables direct agent execution on live Windows system without isolation. Both providers support native Windows input simulation (SendInput API) and clipboard operations. Handles Windows-specific action execution (window management, registry access).
Unique: Implements both Windows Sandbox provider (ephemeral isolated environments with automatic cleanup) and host provider (direct OS execution) with native Windows input simulation (SendInput API) and clipboard support. Handles Windows-specific action execution including window management.
vs alternatives: Windows Sandbox provides better isolation than host execution while avoiding VM overhead; native SendInput API enables more reliable input simulation than generic input methods.
Implements comprehensive telemetry and logging infrastructure capturing agent execution metrics (latency, token usage, action success rate), errors, and performance data. Supports structured logging with contextual information (task ID, agent ID, timestamp). Integrates with external monitoring systems (e.g., Datadog, CloudWatch) for centralized observability. Provides error categorization and automatic error recovery suggestions. Enables debugging through detailed execution logs with configurable verbosity levels.
Unique: Implements structured telemetry and logging system with contextual information (task ID, agent ID, timestamp), error categorization, and automatic error recovery suggestions. Integrates with external monitoring systems for centralized observability.
vs alternatives: More comprehensive than basic logging because it captures metrics and structured context; integration with external monitoring enables centralized observability vs. log file analysis.
Implements the core agent loop (screenshot → LLM reasoning → action execution → repeat) via the ComputerAgent class, with pluggable callback system and custom loop support. Developers can override loop behavior at multiple extension points: custom agent loops (modify reasoning/action selection), custom tools (add domain-specific actions), and callback hooks (inject monitoring/logging). Supports both synchronous and asynchronous execution patterns.
Unique: Provides a callback-based extension system with multiple hook points (pre/post action, loop iteration, error handling) and explicit support for custom agent loop subclassing, allowing developers to override core loop logic without forking the framework. Supports both native computer-use models and composed models with grounding adapters.
vs alternatives: More flexible than frameworks with fixed loop logic; callback system enables non-invasive monitoring/logging vs. requiring loop subclassing, while custom loop support accommodates novel agent architectures that standard loops cannot express.
+7 more capabilities