FGSE Agent - AppWorld Challenge: leaderboard
FGSE Agent - AppWorld Challenge: Task success (%). FGSE arXiv 2611.14326v1: frozen GPT-4o-mini, GPT-4o and DeepSeek-V3.1 ReAct-style agents with FGSE memory updates. Table 1 reports WebArena/AppWorld percentages and tau pass fractions; Table 5 LongMemEval is a fraction, LoCoMo Rouge-L is a percentage. Author FGSE rows are explicit actual experiments; non-FGSE comparison rows with unclear historical/source configuration are held rather than copied. Reliability update rates are not interchangeable with task capability. No backbone fine-tuning.
Metric: Task success (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek-V3.1 (Benchmark Sweep Run: source=fgse_agent_paper; harness=FGSE ReAct; frozen checkpoint; effort=unreported) | 69.4 | |
| 2 | GPT-4o (Benchmark Sweep Run: source=fgse_agent_paper; harness=FGSE ReAct; frozen checkpoint; effort=unreported) | 68.3 | |
| 3 | GPT-4o mini (Benchmark Sweep Run: source=fgse_agent_paper; harness=FGSE ReAct; frozen checkpoint; effort=unreported) | 66.1 |
Interactive version: theaggregate.ai/benchmark?slug=fgse-agent-appworld-challenge · How It Works · Data refreshed daily, snapshot 2026-11-12.