Code Arena has launched its Fullstack capabilities, enabling the evaluation of AI models on complex full-stack web development tasks including multi-step reasoning, tool use, and end-to-end app generation. Kimi K3 (Max) claims the top position, followed by GPT 5.6 Sol (xHigh) at #2 and Claude Fable 5 at #3. This expansion comes as assessments of AI models increasingly reflect their performance in real-world scenarios, with a focus on their agentic capabilities to plan, execute, and refine tasks using structured tool calls, moving beyond simple frontend prototypes to more comprehensive development involving databases and API keys.

Kimi K3: Kimi K3 is a frontier AI model developed by Moonshot AI, recognized for strong performance in coding, agentic workflows, long-horizon reasoning, and visual understanding. In the news, it achieved the top ranking in Code Arena’s new fullstack evaluation category, outperforming other leading models on tasks involving multi-step reasoning and tool use.
Code Arena: Code Arena is a community-driven evaluation platform operated by Arena.ai that benchmarks AI models on coding and software development tasks. It recently expanded its evaluations to include fullstack capabilities, allowing models to act as agents that handle databases, APIs, deployments, and end-to-end app generation using structured tool calls. This update directly enables the current rankings of models on more realistic, real-world software engineering challenges.
GPT 5.6 Sol: GPT 5.6 Sol is OpenAI’s flagship model in the GPT-5.6 series, optimized for complex reasoning, coding, cybersecurity, and agentic tasks with enhanced computer use and design judgment. It secured second place in the updated Code Arena fullstack rankings, reflecting its improved capabilities in executing real-world development workflows compared to prior versions.
Claude Fable 5: Claude Fable 5 is Anthropic’s advanced model focused on ambitious coding projects, large-scale implementations, and autonomous multi-day sessions with robust safeguards. It placed third in Code Arena’s fullstack benchmark, highlighting its competitive edge in tool-assisted, end-to-end software generation amid recent global availability updates.

Model Competition: Recent releases from major AI labs emphasize stronger performance in coding and agentic capabilities, leading to shifting rankings as evaluations evolve to reflect practical development scenarios.
Agentic AI Progress: Frontier models are increasingly evaluated on their ability to function as agents that plan, execute, and refine tasks in real time using structured tool calls for complex software projects.
Benchmark Expansion: Evaluation platforms like Code Arena are extending benchmarks beyond frontend prototypes to fullstack environments that incorporate real databases, API integrations, and deployments.