Ouroboros on OSWorld-Verified: 90.69%, the highest result reported to date

Status: Self-reported result over all 361 tasks. The official per-task scores, prompts, manifests and feasibility records are public here, together with every acting task record that the run produced.

Start here

Read the technical report, Browse all Ouroboros benchmark runs, inspect individual tasks without an API key, or read the public evidence page.

90.69% (327.39 / 361) on OSWorld-Verified with claude-opus-5, screenshot-only, one rollout and a 100-step budget, above the current leaderboard #1 (Intelligence-Indeed, 90.19%). This repository contains every per-task score and prompt, every feasibility record, and each acting task record that the run produced.

Result

Leaderboard figures are read from the official workbook osworld_verified_results.xlsx.

Our companion run with claude-sonnet-4.6, the best published Sonnet 4.6 result (83.27%), is at razzant/ouroboros-osworld-verified-sonnet46.

Per domain

Method

Ouroboros is a general-purpose autonomous agent runtime, not an OSWorld-specific scaffold. The benchmark adapter binds it to the OSWorld VM through a computer-use skill and hands scoring to the official evaluator.

• Task set: evaluation_examples/test_nogdrive.json, all 361 tasks in the standard set excluding the 8 Google Drive tasks. Every task carries a score; none was dropped.
• Observation: screenshot only. No accessibility tree, no set-of-marks.
• Rollouts: 1 (pass@1). One full pass over all 361 tasks on v6.87.0, run as two disjoint batches (103 + 258); every task ran exactly once, and no task was re-run after scoring.
• Step budget: 100. One step = one policy turn = one agent call returning a batch of actions, matching lib_run_single.py, which increments step_idx once per predict() regardless of how many primitive actions that prediction emits. Steps used: median 17, p90 30, max 100. Zero tasks exceeded the cap; the audit is in every task's task_outcome.json under step_budget_audit.
• Feasibility pre-pass: before acting, a separate read-only task inspects the environment and returns INFEASIBLE / PROCEED / UNDETERMINED against a fixed rubric. It cannot modify the VM, its turns count inside the same 100-step budget, and the acting phase may override its verdict. OSWorld-Verified contains 27 infeasible-by-design tasks, and this makes that judgement explicit and auditable.
• Network: tasks whose OSWorld config sets "proxy": true ran through a residential proxy with a distinct session per task.
• Scoring: the official OSWorld evaluator at commit 091f5ef, unmodified. Our only changes to that checkout are in VM provisioning (desktop_env/providers/docker/provider.py: port-lock timeout, and one port scan per VM start instead of four); no task, evaluator or metric was touched.

Contents

<domain>/<task_id>/
  result.txt                 official per-task score
  task_outcome.json          reward, status, step-budget audit
  ouroboros_task_final.json  acting task record; empty when no acting phase ran
  prompt.txt                 the exact prompt for that task
  feasibility_gate.json      pre-pass verdict and its tool trace
  reset_verification.json    evidence the task setup was verified
  task_run_manifest.json     provenance: model, effort, seed commit, dataset commit
results_summary.json         machine-readable scores, per-domain and per-task

The scaffold prompt is published for every task deliberately: it is what lets a reader judge whether the system is engineered against the evaluator or against the task.

Record coverage: score, prompt, manifest, outcome and feasibility files are present for all 361 tasks. The acting task record is non-empty for 338 tasks. It is an empty object for 23 feasibility-only tasks because no acting phase ran.

Not included: OSWorld task definitions (they carry evaluator.expected, i.e. the gold answers) and the evaluator's downloaded reference files. Publishing those alongside trajectories would turn this repository into an answer key.

Provenance and licensing

This is an evidence package produced by the Ouroboros v6.87.0 run identified above. viewer_index.jsonl is a derived index of the unchanged per_task records in results_summary.json; it exists only so the Hugging Face Dataset Viewer can display one stable row per task. The exact evidence snapshot remains f52ebf2248ce0ce0c496db18f5e6edce631304fa.

The task identifiers, benchmark prompts and evaluator outputs originate from OSWorld at the pinned commit. OSWorld is Apache-2.0, as is the original material in this evidence package. The repository does not include OSWorld task definitions, gold answers or downloaded evaluator reference files. See THIRD_PARTY_NOTICES.md for the exact scope and upstream attribution.

Ouroboros

• Source: https://github.com/razzant/ouroboros
• This run: tag v6.87.0, commit 6b831640

Citation

@misc{ouroboros_osworld_opus5,
  title  = {Ouroboros on OSWorld-Verified with Claude Opus 5},
  author = {Razzhigaev, Anton},
  year   = {2026},
  url    = {https://github.com/razzant/ouroboros}
}