FlowerBench: Benchmarking AI Agents on Real Enterprise Work (flower.ai)

The recent launch of FlowerBench introduces a new benchmark for evaluating AI agents on real enterprise workflows, focusing on long-horizon tasks within private organizational environments. This initiative enables AI models to be assessed in context-rich scenarios that leverage proprietary knowledge and tools, although only performance metrics are publicly disclosed. The results from this benchmarking reveal a competitive landscape of AI agents, with OpenAI's Codex (GPT-5.5) at the top, achieving the highest performance with a ranking score of 0.82, followed by Anthropic's Claude Opus models and Google's Gemini, all evaluated on task completion time, token usage, and operational costs. FlowerBench is significant for the AI/ML community as it provides a standardized method to measure the efficacy of AI tools in real-world business applications. The insights drawn from this benchmarking could influence development priorities and investment in AI models, underscoring the importance of operational efficiency alongside technical capabilities. As enterprises increasingly rely on AI to handle complex workflows, tools like FlowerBench could guide researchers and companies in optimizing AI agent deployment and improving their performance across long-term tasks.