OpenGPTEvidence-based model decisions

Search OpenGPT

Models, agents, rankings, and tools

Go toAI model rankingsPublic-source model rankingsGo toAI Agent directoryProducts and comparable public evaluationsGo toChoose a modelStart from your task and build a shortlistGo toTask rankingsBrowse evidence by real-world taskGo toCompare modelsCompare exact versions and key differencesGo toReal-task evaluationBlind-review answers on the same tasksGo toRanking changesSee version and ranking historyGo toModel directoryBrowse official model IDs
Start evaluation

Agent configuration · OSWorld 2.0

OSWorld Agent · MiniMax M3 · Enabled · Standard

The ranked object: model, scaffold, tools, permissions, budget, and environment together.

Compare in this benchmarkOpen source
Binary completion4.6%
Cost per successful taskNot published
Published result rangeNot published

01

Configuration

Agent systemOSWorld Agent · MiniMax M3 · Enabled · Standard
Base modelMiniMax M3
Source model ID or labelMiniMax M3
ScaffoldOSWorld 2.0 computer-use agent
Scaffold versionOSWorld 2.0
ToolsStandard computer-use tool
EnvironmentDesktop and operating-system workflows in reproducible virtual machines
Budget500 steps

02

Outcome

Binary completion4.6%
Partial score22.3%
Cost per successful taskNot published
Reported benchmark cost$258.78
Median completion timeNot published
Human interventionNot published

03

Operation

Repeat-run evidenceNot published
Published result rangeNot published
Safety noteNot published
Comparability grouposworld-2-v2026-06-24-500
Evaluation date
Data captured

04

Trust

EvidenceListed by the benchmark publisher
SourceOSWorld 2.0
Source snapshotTask version v2026.06.24 · 500-step budget
Source data dateNot published by source
Retrieved by OpenGPT
Ranking last changedNo comparable history yet
Last checked
Tasks108 long-horizon computer workflows
Evidence levelPublisher maintained
Evidence note

Rows preserve the reported agent, model, tool mode, task version, step limit, and source metrics.

Open audit ↗

What this result does not prove

This result applies to the exact system and source conditions shown here. It does not establish a universal best agent, production reliability, data governance, or performance on a different benchmark version.

Compare like with like

Only configurations from the same benchmark version and task scope can be placed in one comparison.

OSWorld Agent · Claude Opus 4.8 · Max · Batched tool20.6% · Claude Opus 4.8OSWorld Agent · Claude Opus 4.8 · Max · Standard18.52% · Claude Opus 4.8OSWorld Agent · Claude Opus 4.7 · Max · Batched tool18.2% · Claude Opus 4.7OSWorld Agent · Claude Opus 4.7 · Max · Standard13.9% · Claude Opus 4.7

Source-reported data; missing values remain unknown.

Validate before adoption

Public rankings create a shortlist. A small private trial decides whether the configuration fits your work.

  1. 1

    Choose 10–20 representative tasks and define a clear pass condition.

  2. 2

    Lock the agent, model, tools, permissions, and budget before testing.

  3. 3

    Run each important task more than once; record success, cost, time, and takeovers.

  4. 4

    Review failures and data-handling risks before a production decision.

← All AI Agent rankings