Research map / Large language models / Agents, code and math
EffGen: Enabling Small Language Models as Capable Autonomous Agents
Gaurav Srivastava, Aafiya Hussain, Chi Wang, Yingyan Celine Lin, Xuan Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add prompt optimization (ablation) | EffGen · 13-benchmark average | average accuracy: 36.21 -> 47.44 (1.5B) | Improved |
| add prompt optimization (ablation) | EffGen Qwen2.5-32B · 13-benchmark average | average accuracy: 68.54 -> 70.97 | Improved |
| add complexity routing (ablation) | EffGen Qwen2.5-1.5B · 13-benchmark average | average accuracy: 43.87 -> 47.44 | Improved |
| add complexity routing (ablation) | EffGen Qwen2.5-32B · 13-benchmark average | average accuracy: 63.12 -> 70.97 | Improved |
| add task decomposition (ablation) | EffGen Qwen2.5-7B · 13-benchmark average | average accuracy: 58.42 -> 63.07 | Improved |
| add three-tier memory system (ablation) | EffGen Qwen2.5-1.5B · 13-benchmark average | average accuracy: 45.68 -> 47.44 | Improved |
| EffGen framework versus raw ReAct (all components) (ablation) | Qwen2.5-1.5B · 13-benchmark average | average accuracy: 34.28 -> 47.44 | Improved |
| use LangChain instead of raw model | Qwen2.5-1.5B · 13-benchmark average | average accuracy: 34.28 -> 32.81 | Hurt |
| use agentic frameworks on GSM8K (large model) | Qwen2.5-32B · GSM8K | accuracy: 95.60 -> 95.75 | No clear effect |
| give raw model direct search tool access (Raw+Search) (ablation) | Qwen2.5-1.5B · SimpleQA | accuracy: 4% -> 8% (EffGen 40%) | Improved |
| add standard retrieval frameworks (LangChain) | Qwen2.5-1.5B · ARC-Challenge | accuracy: 67.24 -> 59.13 | Hurt |
| add EffGen filtering of retrieved content | Qwen2.5-1.5B · ARC-Challenge | accuracy: 67.24 -> 78.34 | Improved |
Closest papers by meaning
- PACE: Two-Timescale Self-Evolution for Small Language Model Agents (NeurIPS 2026 · similarity 0.87)
- The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution (ICLR 2026 · similarity 0.86)
- MAS$^2$: Self-Generative, Self-Configuring, Self-Rectifying Multi-Agent Systems (ICLR 2026 · similarity 0.86)
- WebArena: A Realistic Web Environment for Building Autonomous Agents (ICLR 2024 · similarity 0.86)
- ShortcutsBench: A Large-Scale Real-world Benchmark for API-based Agents (ICLR 2025 · similarity 0.85)
- FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents (ICML 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
