Reference figures and datasets from sipi.bot on AI agent spending patterns, cost-per-task benchmarks, provider price comparisons, and runaway incident analysis. All data is free to cite under CC BY 4.0.
We publish reference figures because the market for autonomous agents is growing faster than the public data about it. Teams are deploying agents that can spend money — on LLM APIs, cloud compute, SaaS services, ads, and tool-call fees — but there is almost no open data on what normal agent spending looks like, what incidents cost, or where the biggest waste happens.
Our research covers the gaps: cost-per-task benchmarks across 15 common agent workflows (code generation, research, data processing, customer support), spending scenarios modelled from published provider pricing, runway agent incident documentation, and provider-level cost comparisons. Every report ships with raw data in CSV and JSON formats so you can run your own analysis.
Cross-provider analysis of LLM API spending patterns across 15 common agent workflow tasks. Identifies the specific patterns that drive cost overruns — retry loops, suboptimal model choices, unused provisioned throughput — and quantifies the savings from fixing each one.
Our research combines three sources: public incident reports from engineering teams who publish postmortems, and published provider pricing and documentation. All reports are timestamped and versioned so citations can reference a specific data snapshot.
All research reports and datasets are licensed under CC BY 4.0. You may cite, reproduce, and adapt the data for any purpose — including commercial use — as long as you provide attribution to sipi.bot. If you publish work based on these figures, we would appreciate a link back, though this is not a requirement of the license.