Agentic & Tool Use benchmark

# OSWorld leaderboard

> OSWorld results for 8 AI models, led by Claude Sonnet 4.6 at 72.1%. What the benchmark measures, who runs it, and a source for every score.
- Canonical page: https://noometry.com/benchmarks/osworld
- Last updated: 2026-10-10
- Title: OSWorld Leaderboard (October 2026): Scores by Model

As of October 2026, Claude Sonnet 4.6 has the highest published OSWorld score on Noometry at 72.1%, out of 8 models with results.

Last verified October 10, 2026

## About OSWorld

Open-ended computer-use tasks in real desktop operating systems and applications.

- **Category:** [Agentic & Tool Use](https://noometry.com/best/agentic)
- **Introduced:** 2024
- **Size:** 369 tasks
- **Format:** Computer use
- **Unit:** Percent (random guessing ≈ 0%)
- **Official site:** [os-world.github.io](https://os-world.github.io)

## Top 8 models

Top models on OSWorld

1.  Claude Sonnet 4.6 72.1%
2.  Claude Opus 4.5 66.3%
3.  Kimi K2.5 63.3%
4.  Claude Sonnet 4.5 62.9%
5.  Claude Sonnet 4 43.9%
6.  Claude 3.7 Sonnet 35.8%
7.  o3 23%
8.  Qwen2.5-VL 72B Instruct 5%
9.  020406080

Sponsored placements are available on pages like this one. [Advertise on Noometry](https://noometry.com/advertise)

## All results

OSWorld results by model
| # | Model | Provider | Score | Setting | Source | Date |
| --- | --- | --- | --- | --- | --- | --- |
| 1 | [Claude Sonnet 4.6](https://noometry.com/models/claude-sonnet-4-6) | [Anthropic](https://noometry.com/providers/anthropic) | 72.1% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 2 | [Claude Opus 4.5](https://noometry.com/models/claude-opus-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 66.3% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 3 | [Kimi K2.5](https://noometry.com/models/kimi-k2-5) | [Moonshot AI](https://noometry.com/providers/moonshot) | 63.3% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 4 | [Claude Sonnet 4.5](https://noometry.com/models/claude-sonnet-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 62.9% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 5 | [Claude Sonnet 4](https://noometry.com/models/claude-sonnet-4) | [Anthropic](https://noometry.com/providers/anthropic) | 43.9% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 6 | [Claude 3.7 Sonnet](https://noometry.com/models/claude-3-7-sonnet) | [Anthropic](https://noometry.com/providers/anthropic) | 35.8% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 7 | [o3](https://noometry.com/models/o3) | [OpenAI](https://noometry.com/providers/openai) | 23% | medium | [Epoch AI](https://epoch.ai/benchmarks) |  |
| 8 | [Qwen2.5-VL 72B Instruct](https://noometry.com/models/qwen2-5-vl-72b-instruct) |  [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 5% |  | [Epoch AI](https://epoch.ai/benchmarks) |  |

## Compare the leaders

-   [Claude Sonnet 4.6 vs Claude Opus 4.5](https://noometry.com/compare/claude-opus-4-5-vs-claude-sonnet-4-6)
-   [Claude Sonnet 4.6 vs Kimi K2.5](https://noometry.com/compare/claude-sonnet-4-6-vs-kimi-k2-5)
-   [Claude Sonnet 4.6 vs Claude Sonnet 4.5](https://noometry.com/compare/claude-sonnet-4-5-vs-claude-sonnet-4-6)
-   [Claude Sonnet 4.6 vs Claude Sonnet 4](https://noometry.com/compare/claude-sonnet-4-vs-claude-sonnet-4-6)
-   [Claude Opus 4.5 vs Kimi K2.5](https://noometry.com/compare/claude-opus-4-5-vs-kimi-k2-5)
-   [Claude Opus 4.5 vs Claude Sonnet 4.5](https://noometry.com/compare/claude-opus-4-5-vs-claude-sonnet-4-5)

## Other agentic & tool use benchmarks

-   [Terminal-Bench](https://noometry.com/benchmarks/terminal-bench)
-   [APEX-Agents](https://noometry.com/benchmarks/apex-agents)
-   [Berkeley Function Calling Leaderboard](https://noometry.com/benchmarks/bfcl)
-   [OSWorld 2.0](https://noometry.com/benchmarks/osworld-2)
-   [GDPval](https://noometry.com/benchmarks/gdpval)
-   [Remote Labor Index](https://noometry.com/benchmarks/remote-labor-index)
-   [TheAgentCompany](https://noometry.com/benchmarks/the-agent-company)
-   [τ²-bench Airline](https://noometry.com/benchmarks/tau2-airline)
-   [τ²-bench Banking](https://noometry.com/benchmarks/tau2-banking)
-   [τ²-bench Retail](https://noometry.com/benchmarks/tau2-retail)
-   [τ²-bench Telecom](https://noometry.com/benchmarks/tau2-telecom)
-   [Cybench](https://noometry.com/benchmarks/cybench)

## Frequently asked questions

### What does OSWorld measure?

Open-ended computer-use tasks in real desktop operating systems and applications.

### Which model has the highest OSWorld score?

As of October 2026, Claude Sonnet 4.6 has the highest published OSWorld score on Noometry at 72.1%, out of 8 models with results.

### What is the best open-weight model on OSWorld?

Kimi K2.5 has the highest OSWorld accuracy among open-weight models at 63.3%, ranking 3 of 8 overall.

### Cite this page

Noometry. (2026). OSWorld leaderboard. Retrieved October 10, 2026, from https://noometry.com/benchmarks/osworld

Quote Noometry with a link back to this page. It is also available in [Markdown](https://noometry.com/md/benchmarks/osworld.md).