TASK MATRIX5 CUJs Evaluated
Task & CUJ Explorer
Explore cross-model performance across authentic Flutter and Dart developer workflows. Click any cell in the matrix to inspect the full trial execution, rubric, trajectory, and logs.
Cross-Model Performance Heatmap
Color scaled by reward (Green ≥ 0.80, Yellow 0.50–0.79, Red < 0.50). Errored trials hatched in gray.| Critical User Journey (CUJ) | Category |
Claude 3.7 Sonnet
claude-code
|
Gemini 3.5 Pro
antigravity-sdk
|
o3
codex-agent
|
GPT 5
codex-agent
|
DeepSeek R1
deepseek-cli
|
Gemini 3.5 Flash
antigravity-sdk
|
Claude 3.5 Sonnetclaude-code
|
GPT 4ocodex-agent
|
DeepSeek V3deepseek-cli
|
DeepSeek Coder V2deepseek-cli
|
Claude 3.5 Haikuclaude-code
|
GPT 4o Minicodex-agent
|
Gemini 3.1 Flash Litegemini-cli
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Custom RenderObject & Canvas | Advanced Rendering | 0.71 | 0.69 | 0.73 | 0.72 | 0.72 | 0.65 | 0.52 | 0.47 | 0.49 | 0.42 | 0.36 | AgentTimeoutError | AgentTimeoutError |
| Build Command-Line CLI App | Dart Utilities | 0.97 | 0.98 | 0.97 | 0.98 | 0.99 | 0.93 | 0.79 | 0.71 | 0.71 | 0.61 | 0.56 | 0.48 | 0.39 |
| Offline SQLite Sync Repository | Data & Storage | 0.93 | 0.91 | 0.90 | 0.86 | 0.87 | 0.80 | 0.71 | 0.65 | 0.64 | 0.55 | 0.48 | 0.39 | 0.33 |
| Manage State with BLoC | State Management | 0.98 | 0.99 | 0.97 | 0.97 | 0.94 | 0.88 | 0.75 | 0.69 | 0.67 | 0.59 | 0.55 | 0.46 | AgentTimeoutError |
| Adaptive Material & Cupertino UI | Multi-Platform UI | 0.85 | 0.87 | 0.85 | 0.84 | 0.78 | 0.78 | 0.61 | 0.60 | 0.59 | 0.50 | 0.45 | 0.39 | 0.28 |
Model Strengths & Weaknesses Across CUJs
Claude 3.7 Sonnet
claude-code
Strongest CUJs:
- Manage State with BLoC (0.98)
- Build Command-Line CLI App (0.97)
- Offline SQLite Sync Repository (0.93)
Growth Areas / Weaknesses:
None recorded.
Gemini 3.5 Pro
antigravity-sdk
Strongest CUJs:
- Manage State with BLoC (0.99)
- Build Command-Line CLI App (0.98)
- Offline SQLite Sync Repository (0.91)
Growth Areas / Weaknesses:
None recorded.
o3
codex-agent
Strongest CUJs:
- Build Command-Line CLI App (0.97)
- Manage State with BLoC (0.97)
- Offline SQLite Sync Repository (0.90)
Growth Areas / Weaknesses:
None recorded.
GPT 5
codex-agent
Strongest CUJs:
- Build Command-Line CLI App (0.98)
- Manage State with BLoC (0.97)
- Offline SQLite Sync Repository (0.86)
Growth Areas / Weaknesses:
None recorded.
DeepSeek R1
deepseek-cli
Strongest CUJs:
- Build Command-Line CLI App (0.99)
- Manage State with BLoC (0.94)
- Offline SQLite Sync Repository (0.87)
Growth Areas / Weaknesses:
None recorded.
Gemini 3.5 Flash
antigravity-sdk
Strongest CUJs:
- Build Command-Line CLI App (0.93)
- Manage State with BLoC (0.88)
- Offline SQLite Sync Repository (0.80)
Growth Areas / Weaknesses:
None recorded.
Claude 3.5 Sonnet
claude-code
Strongest CUJs:
- Build Command-Line CLI App (0.79)
- Manage State with BLoC (0.75)
- Offline SQLite Sync Repository (0.71)
Growth Areas / Weaknesses:
None recorded.
GPT 4o
codex-agent
Strongest CUJs:
- Build Command-Line CLI App (0.71)
- Manage State with BLoC (0.69)
- Offline SQLite Sync Repository (0.65)
Growth Areas / Weaknesses:
- Custom RenderObject & Canvas (0.47)
DeepSeek V3
deepseek-cli
Strongest CUJs:
- Build Command-Line CLI App (0.71)
- Manage State with BLoC (0.67)
- Offline SQLite Sync Repository (0.64)
Growth Areas / Weaknesses:
- Custom RenderObject & Canvas (0.49)
DeepSeek Coder V2
deepseek-cli
Strongest CUJs:
- Build Command-Line CLI App (0.61)
- Manage State with BLoC (0.59)
- Offline SQLite Sync Repository (0.55)
Growth Areas / Weaknesses:
- Custom RenderObject & Canvas (0.42)
Claude 3.5 Haiku
claude-code
Strongest CUJs:
- Build Command-Line CLI App (0.56)
- Manage State with BLoC (0.55)
Growth Areas / Weaknesses:
- Custom RenderObject & Canvas (0.36)
- Adaptive Material & Cupertino UI (0.45)
- Offline SQLite Sync Repository (0.48)
GPT 4o Mini
codex-agent
Strongest CUJs:
No high-scoring tasks.
Growth Areas / Weaknesses:
- Custom RenderObject & Canvas (Error)
- Adaptive Material & Cupertino UI (0.39)
- Offline SQLite Sync Repository (0.39)
Gemini 3.1 Flash Lite
gemini-cli
Strongest CUJs:
No high-scoring tasks.
Growth Areas / Weaknesses:
- Manage State with BLoC (Error)
- Custom RenderObject & Canvas (Error)
- Adaptive Material & Cupertino UI (0.28)
Evaluated Critical User Journeys
Each task is grounded in realistic development workflows and includes automated grading suites:
Advanced Rendering
Custom RenderObject & Canvas
Implement a custom RenderBox with layout constraints, custom painting, intrinsic dimensions, and pointer hit-testing.
Dart Utilities
Build Command-Line CLI App
Create a robust Dart command-line interface application with argument parsing, formatted output, and exit code handling.
Data & Storage
Offline SQLite Sync Repository
Build an offline-first repository using SQLite with background synchronization, retry queues, and conflict resolution.
State Management
Manage State with BLoC
Implement an immutable state management layer using package:flutter_bloc, connecting UI events to business logic with unit tests.
Multi-Platform UI
Adaptive Material & Cupertino UI
Build adaptive Flutter widgets that render Material 3 on Android/Web and native Cupertino patterns on iOS/macOS.