# OSWorld 2.1 Offline leaderboard

Agentic benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelchoice.dev/leaderboards/benchmarks/agentic-os-world21-offline

OSWorld 2.1 offline subset (82 tasks). Every model the catalog carries a published OSWorld 2.1 Offline value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Agentic |
| Measure | Interactive computer-use evaluation; partial score |
| Tasks | 82 offline computer-use workflows |
| Difficulty | Long-horizon professional workflows |
| Published by | [OSWorld 2.1 release (xlang-ai/OSWorld-V2, tag osworld-v2.1)](https://github.com/xlang-ai/OSWorld-V2/releases) |

The official offline subset of OSWorld 2.1: 82 of the 108 long-horizon computer-use workflows, run with no internet access in the agent's virtual machine.

## OSWorld 2.1 Offline ranking

11 models with a published OSWorld 2.1 Offline value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Interactive computer-use evaluation; partial score | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Claude Fable 5.2](https://modelchoice.dev/models/claude-fable-5-2) | Anthropic | 89.3 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Claude Opus 5.5](https://modelchoice.dev/models/claude-opus-5-5) | Anthropic | 87.2 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Claude Sonnet 5.5](https://modelchoice.dev/models/claude-sonnet-5-5) | Anthropic | 83.9 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Claude Fable 5.1](https://modelchoice.dev/models/claude-fable-5-1) | Anthropic | 83.5 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Gemini 4 Carbon](https://modelchoice.dev/models/gemini-4-carbon) | Google | 81.5 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Gemini 4 Hydrogen](https://modelchoice.dev/models/gemini-4-hydrogen) | Google | 77.6 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Gemini 4 Argon](https://modelchoice.dev/models/gemini-4-argon) | Google | 74.7 | Observed 2026-11-21 · source benchlm:models |
| 8 | [GPT-6.1 Terra](https://modelchoice.dev/models/gpt-6-1-terra) | OpenAI | 74.2 | Observed 2026-11-21 · source benchlm:models |
| 9 | [Claude Haiku 5.5](https://modelchoice.dev/models/claude-haiku-5-5) | Anthropic | 72.4 | Observed 2026-11-21 · source benchlm:models |
| 10 | [GPT-6 Luna](https://modelchoice.dev/models/gpt-6-luna) | OpenAI | 48.9 | Observed 2026-11-21 · source benchlm:models |
| 11 | [Claude Haiku 4.5](https://modelchoice.dev/models/claude-haiku-4-5) | Anthropic | 15.7 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
