Symbolica Agentica SDK: Open-Source Agent Framework Scores 36% on ARC-AGI-3
Symbolica has released the Agentica SDK, an open-source agent framework that achieved 36.08% on ARC-AGI-3 — the new benchmark where frontier LLMs score below 1%. For comparison, Claude Opus 4.6 Max scores 0.25% at $8,900, while Agentica reaches 36.08% for $1,005.
The Agentica SDK is built on the premise that code is the most expressive interface through which models can interact with their environment. Rather than fixed tool definitions, agents using Agentica can integrate with live code — functions, classes, objects, and entire SDKs — as first-class tools. The framework is available for both Python and TypeScript.
On ARC-AGI-3, the Agentica-powered agent (Arcgentica) passed 113 out of 182 playable levels and completed 7 of 25 available games. The system uses REPL (Read-Eval-Print Loop) agents that write and execute code to solve abstract reasoning tasks, demonstrating that agent architectures can dramatically outperform raw model capability on novel problems.
The SDK, agent harness, and ARC-AGI-3 solution are all open-source on GitHub: https://github.com/symbolica-ai/agentica-python-sdk and https://github.com/symbolica-ai/arcgentica.
Agentica demonstrates that the gap between model intelligence and agent intelligence is widening — the right agent architecture can amplify a model's effective capability by orders of magnitude on reasoning tasks.
← Back to all articles
The Agentica SDK is built on the premise that code is the most expressive interface through which models can interact with their environment. Rather than fixed tool definitions, agents using Agentica can integrate with live code — functions, classes, objects, and entire SDKs — as first-class tools. The framework is available for both Python and TypeScript.
On ARC-AGI-3, the Agentica-powered agent (Arcgentica) passed 113 out of 182 playable levels and completed 7 of 25 available games. The system uses REPL (Read-Eval-Print Loop) agents that write and execute code to solve abstract reasoning tasks, demonstrating that agent architectures can dramatically outperform raw model capability on novel problems.
The SDK, agent harness, and ARC-AGI-3 solution are all open-source on GitHub: https://github.com/symbolica-ai/agentica-python-sdk and https://github.com/symbolica-ai/arcgentica.
Agentica demonstrates that the gap between model intelligence and agent intelligence is widening — the right agent architecture can amplify a model's effective capability by orders of magnitude on reasoning tasks.
Comments