DIVE: Scaling Diversity in Agentic Task Synthesis for Generalizable Tool Use
About
Recent work synthesizes agentic tasks for post-training tool-using LLMs, yet robust generalization under shifts in tasks and toolsets remains an open challenge. We trace this brittleness to insufficient diversity in synthesized tasks. Scaling diversity is difficult because training requires tasks to remain executable and verifiable, while generalization demands coverage of diverse tool types, toolset combinations, and heterogeneous tool-use patterns. We propose DIVE, an evidence-driven recipe that inverts synthesis order, executing diverse, real-world tools first and reverse-deriving tasks strictly entailed by the resulting traces, thereby providing grounding by construction. DIVE scales structural diversity along two controllable axes, tool-pool coverage and per-task toolset variety, and an Evidence Collection--Task Derivation loop further induces rich multi-step tool-use patterns across 373 tools in five domains. Training Qwen3-8B on DIVE data (48k SFT + 3.2k RL) improves by +22 average points across 9 OOD benchmarks and outperforms the strongest 8B baseline by +68. Remarkably, controlled scaling analysis reveals that diversity scaling consistently outperforms quantity scaling for OOD generalization, even with 4x less data.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Medical Agent Task Execution | MedAgentBench | Success Rate57.3 | 24 | |
| Domain Deep Research Tool Use | FinSearchComp Global-T2 | Success Rate67.3 | 12 | |
| Domain Deep Research Tool Use | FinSearchComp Global-T3 | Success Rate37.3 | 12 | |
| In-distribution Tool Use | DIVE-Eval | Success Rate42.5 | 12 | |
| Financial Specialist Tool Use | Finance Agent Benchmark | Success Rate34 | 12 | |
| General Deep Research Tool Use | GAIA | Success Rate61.2 | 12 | |
| General Deep Research Tool Use | Browsecomp | Success Rate16.4 | 12 | |
| General Deep Research Tool Use | xbench DeepSearch | Success Rate58.1 | 12 | |
| General Deep Research Tool Use | HLE | Success Rate17.8 | 12 | |
| Zero-Shot Generalist Tool Use | Toolathlon | Success Rate8.3 | 12 |