SaaS-Bench: Results as Published

SaaS-Bench tests whether a computer-use agent can complete professional workflows by operating real, self-hosted SaaS applications through a browser. The benchmark has 106 tasks across 23 SaaS systems in six domains (business, healthcare, software, teamwork, agriculture and media), split into a text-only track and a multimodal track. Each task is a multi-step business workflow, and a per-task verifier reads the running application's database, API and page state, plus LLM-judged content checks, to give weighted checkpoint scores with partial credit. The paper also reports strict end-to-end task completion, which is much lower.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.