Sarms paper claims GraphWorkflow runs compiled agent graphs up to 62.5x faster than LangGraph
KyeGomezB · x · 2026-09-08
The Swarms team published a systems paper on GraphWorkflow, the graph execution engine inside the open-source Swarms framework, built on a compile-once, execute-many model: an optimized execution plan is built up front, avoiding the repeated cost of interpreting and constructing large agent graphs.
Key results vs LangGraph 1.0.4, across 15 topology/size configurations (10–200 nodes), medians of 9 samples with 95% confidence intervals:
- 7.0x geometric-mean speedup on compiled graph execution, up to 62.5x on 200-node deep chains
- 21.6x–31.3x faster compilation
- 7.9x faster on the cold build-compile-execute path
The paper includes a formal execution model with proofs, a cost taxonomy for agent orchestration, engine design details, a programming-model comparison with LangGraph, and full benchmarks. The DAG programming model supports parallel execution, branching, fan-in/fan-out, loops, and dependencies. The benchmark harness is open, and every number is claimed to be reproducible.
More from coding & agent
- AXL launches remote MCP server exposing 1,284 sales and marketing APIs to Claude and Codex — EdTech_Padre · 2026-10-27
- GitHub's Thomas Dohmke ships Entire, a Git forge built for agent-human collaboration — carrycooldude · 2026-10-27
- OmarchyTV hits GitHub: Omarchy-style TV launcher with AI TUI and MCP — alexcovo_eth · 2026-10-27
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27
- Reddit Thread: Keeping AI Agents Grounded in Current Project Information — Few-Ad-1358 · 2026-10-27
- He gave his AI agent a phone number and email — then hired it as social media manager — heyshrutimishra · 2026-10-27