Tests whether an agent can produce legal deliverables that meet every requirement without material hallucinations, helping you assess how much drafting and document work you could delegate before professional review. These results use the private 120-task v1.1 evaluation, not earlier LAB scoring.
| Model / creator | Version | Reported score |
|---|---|---|
 Grok 4.7xAI · Proprietary | 1.1 / private 120 / hallucination-gated | 9.4% |
 Muse Spark 1.3Meta · Proprietary | 1.1 / private 120 / hallucination-gated | 8.9% |
 GPT-6 AstraOpenAI · Proprietary | 1.1 / private 120 / hallucination-gated | 8.6% |
 Claude Fable 5.2Anthropic · Proprietary | 1.1 / private 120 / hallucination-gated | 7.2% |
| GPT-6.1 SolOpenAI · Proprietary | 1.1 / private 120 / hallucination-gated | 6.9% |
| Claude Fable 5.1Anthropic · Proprietary | 1.1 / private 120 / hallucination-gated | 6.4% |
 Kimi K3Moonshot AI · Open source / open weights | 1.1 / private 120 / hallucination-gated | 5.3% |
| Claude Opus 5.5Anthropic · Proprietary | 1.1 / private 120 / hallucination-gated | 4.2% |
effort: max · harness: Stirrup · 120 private tasks; v1.1 three-judge hallucination-gated all-pass rate. Distinct from ungated 10.8% Fable all-pass and earlier v1.0. · Tools enabled
Independent evaluation via Anthropic system card ↗effort: max · harness: Stirrup · 120 private tasks; v1.1 three-judge hallucination-gated all-pass rate. Distinct from ungated 10.8% Fable all-pass and earlier v1.0. · Tools enabled
Independent evaluation via Anthropic system card ↗effort: max · harness: Stirrup · 120 private tasks; v1.1 three-judge hallucination-gated all-pass rate. Distinct from ungated 10.8% Fable all-pass and earlier v1.0. · Tools enabled
Independent evaluation via Anthropic system card ↗effort: max · harness: Stirrup · 120 private tasks; v1.1 three-judge hallucination-gated all-pass rate. Distinct from ungated 10.8% Fable all-pass and earlier v1.0. · Tools enabled
Independent evaluation via Anthropic system card ↗effort: max · harness: Stirrup · 120 private tasks; v1.1 three-judge hallucination-gated all-pass rate. Distinct from ungated 10.8% Fable all-pass and earlier v1.0. · Tools enabled
Independent evaluation via Anthropic system card ↗effort: xhigh · harness: Stirrup · October 8 published v1.1 snapshot; hallucination-gated all-pass rate; not an inferred latest leaderboard value. · Tools enabled
effort: max · harness: Stirrup · October 8 published v1.1 snapshot; hallucination-gated all-pass rate; not an inferred latest leaderboard value. · Tools enabled
effort: max · harness: Stirrup · October 8 published v1.1 snapshot; hallucination-gated all-pass rate; not an inferred latest leaderboard value. · Tools enabled
Tests rebuilding parts, assemblies and circuit boards from drawings and images, helping you assess AI support for client design and prototyping work.
Gaia2Tests whether an assistant can coordinate work across email, calendars, shopping and other apps, helping you assess how much routine planning and execution you could hand off.
OSWorldTests whether an agent can carry out work through desktop applications, helping you assess which computer tasks you might delegate instead of doing every click yourself.
