# SaaS-Bench ![Category](https://img.shields.io/badge/Category-SaaS%20Product%20Agent-1f6feb) ![Metric](https://img.shields.io/badge/Metric-%25%20Success-2da44e) ![Saturation](https://img.shields.io/badge/Saturation-Low-d29922) ![Created](https://img.shields.io/badge/Created-2026-555) > **Quick answer:** SaaS-Bench tests AI agents on realistic workflows within common business SaaS web applications — CRMs, project-management tools, and similar products — evaluating whether an agent can navigate and operate these tools as a real user would. Kimi K3 scores 60.1% as of July 2026. ## At a Glance **What it tests:** An agent's ability to navigate and complete tasks within real SaaS web applications, mirroring how a human employee would use business software. **Why it matters:** SaaS automation is one of the largest near-term commercial opportunities for AI agents. SaaS-Bench measures readiness for this specific, high-value deployment category. **Known limitations:** As an emerging benchmark, the specific set of SaaS applications tested and task-sourcing methodology are not yet independently published outside its citation by Moonshot AI. ## What SaaS-Bench Measures SaaS-Bench evaluates an agent's ability to complete realistic tasks within common business SaaS applications — such as configuring settings, entering and retrieving data, and completing multi-step workflows across app UIs. This complements broader GUI-agent benchmarks (like OSWorld) by focusing specifically on business-software usage patterns. ## Benchmark Specifications | Field | Value | |---|---| | Task category | Agent / SaaS product usage | | Metric | % task success | | Saturation | Low | | Created by | Not yet independently documented | ## How SaaS-Bench Is Scored Agents attempt tasks within SaaS applications, with outcomes verified against the expected application state, producing an aggregate % task success rate. ## State-of-the-Art Results | Rank | Model | Score | Source | Date | |---|---|---|---|---| | 1 | [Kimi K3](/models/moonshot/kimi-k3) | 60.1% | [Kimi K3 technical report](https://www.kimi.com/blog/kimi-k3) | 2026-07 | *Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.* ## SaaS-Bench on Benchgen > No Benchgen results yet — [be the first to run SaaS-Bench](/benchmarks/platform/saas-bench). ## SaaS-Bench vs Other Benchmarks | Benchmark | What it tests | Saturation | |---|---|---| | **SaaS-Bench** | SaaS product usage workflows | Low | | [OSWorld 2.0](/benchmarks/osworld/osworld-20) | Expanded, harder GUI agent workflows | Low | | [AutomationBench](/benchmarks/platform/automationbench) | Business workflow automation | Low | | [MCP Atlas](/benchmarks/scale-ai/mcp-atlas) | MCP tool-use agentic workflows | Low | ## Run SaaS-Bench on Your Model Benchgen lets you run SaaS-Bench against your own agent, tracking SaaS product usage task success rates before rolling out agentic automation to business software. ## Frequently Asked Questions

What is SaaS-Bench?

SaaS-Bench is a benchmark testing AI agents on realistic workflows within common business SaaS web applications, such as CRMs and project-management tools.

What does a good score look like on SaaS-Bench?

Kimi K3 reports 60.1% as of July 2026, a solid frontier-level result reflecting reasonable reliability navigating real business software.

Who created SaaS-Bench?

SaaS-Bench's originating team is not yet independently documented outside of its citation in Kimi K3's July 2026 technical report.

SaaS-Bench — Results

SaaS-Bench — Evaluations