WikiTrace Agent Evaluation - ScienceWorld: leaderboard
WikiTrace Agent Evaluation - ScienceWorld: Three-trial mean native task success (%). WikiTrace arXiv 2611.14320v1 Tables 1-2: mean task success over three trials. All six memory configurations remain native evaluation runs, not new catalog checkpoints. The main model and decoding are bound to the primary protocol below; no current-repository default is inferred. Three GLM suites in Table 2 are separate from the eight-suite main average.
Metric: Three-trial mean native task success (%). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash (High) | 83.7 | #254 (DeepSeek V4 Flash) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=wikitrace-agent-evaluation-scienceworld · How It Works · Data refreshed daily, snapshot 2026-11-12.