# Toolathlon-Verified ![Category](https://img.shields.io/badge/Category-Tool--Use%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:** Toolathlon-Verified is a verified benchmark testing AI agents on multi-tool, multi-step workflows that require chaining together diverse tools to complete a task — an "athlon" of sequential tool-use challenges. Kimi K3 scores 76.5% as of July 2026. ## At a Glance **What it tests:** An agent's ability to select, sequence, and correctly invoke multiple different tools across a multi-step workflow to reach a verified final outcome. **Why it matters:** Real-world agent deployments often require juggling many tools (search, code execution, file systems, APIs). Toolathlon-Verified specifically stresses this multi-tool coordination capability. **Known limitations:** As an emerging benchmark, exact tool inventory and task composition are not yet independently published outside its citation by Moonshot AI. ## What Toolathlon-Verified Measures Toolathlon-Verified evaluates an agent's ability to coordinate multiple distinct tools across a sequential, multi-step workflow, with task outcomes verified programmatically. The "Verified" designation indicates that task setups and success criteria have been validated/corrected, similar to the "Verified" convention used in benchmarks like SWE-Bench Verified and OSWorld-Verified. ## Benchmark Specifications | Field | Value | |---|---| | Task category | Agent / multi-tool workflows | | Metric | % task success | | Saturation | Low | | Created by | Not yet independently documented | ## How Toolathlon-Verified Is Scored Agents complete multi-tool workflows, with final outcomes verified against expected results, producing an aggregate % task success rate. ## State-of-the-Art Results | Rank | Model | Score | Source | Date | |---|---|---|---|---| | 1 | [Kimi K3](/models/moonshot-ai/kimi-k3) | 76.5% | [Kimi K3 technical report](https://www.kimi.com/blog/kimi-k3) | 2026-07 | *Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.* ## Toolathlon-Verified on Benchgen > No Benchgen results yet — [be the first to run Toolathlon-Verified](/benchmarks/hkust-nlp/toolathlon-verified). ## Toolathlon-Verified vs Other Benchmarks | Benchmark | What it tests | Saturation | |---|---|---| | **Toolathlon-Verified** | Multi-tool, multi-step agentic workflows | Low | | [MCP Atlas](/benchmarks/scale-ai/mcp-atlas) | MCP tool-use agentic workflows | Low | | [MCPMark-Verified](/benchmarks/wu-et-al/mcpmark-verified) | MCP protocol tool-use benchmark | Low | | [BFCL v3](/benchmarks/berkeley/bfcl-v3) | Function-calling accuracy | Medium | ## Run Toolathlon-Verified on Your Model Benchgen lets you run Toolathlon-Verified against your own agent harness, tracking multi-tool task success rates as you expand your agent's tool inventory. ## Frequently Asked Questions

What is Toolathlon-Verified?

Toolathlon-Verified is a benchmark testing AI agents on multi-tool, multi-step workflows requiring coordination across diverse tools, with verified task setups and success criteria.

What does a good score look like on Toolathlon-Verified?

Kimi K3 reports 76.5% as of July 2026, a strong result for multi-tool workflow coordination.

Who created Toolathlon-Verified?

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

Rank	Model	Score
1	mimo-v2-6-pro-rl	76.9
2	kimi-k3	76.5
3	deepseek-v4-flash-vision-exp	75.9
4	nex-n2-5-max	74.7
5	deepseek-v4-pro	74.1
6	hy4-preview	74.1
7	mimo-v2-6-flash-rl	73.6
8	qwen3-8-flash-next	73.5
9	glm-5-3	73
10	muse-spark-open-816b-a42b	72.8
11	qwen3-8-max	72.5
12	ornith-1-5-397b	71.2
13	nanbeige4-3-33b	70.4
14	nex-n2-5-pro	68.5
15	k2-horizon-375b-a23b	65.3
16	ornith-1-5-35b-a3b	48.7
17	ornith-1-5-9b	41.2

Toolathlon-Verified — Results

Toolathlon-Verified — Phases

Toolathlon-Verified