AgentReducto
a benchmark suite and a family of fine-tuned models specializing in the Reducto API.
Benchmark
20 Models ×
22 Hard Probes
22 Hard Probes
432 runs across frontier models — Kimi, Claude, GPT, Gemini, Qwen, and more — on API-specific tasks no general benchmark covers.
View results →
Live Demo
Try the Model
in Your Browser
in Your Browser
The quantized 0.8B model running in-browser via WebGPU. Throw hard multi-hop document tasks at it — extract, split, classify, chain.
Open demo →
MCP Server
Reducto for
Claude & Cursor
Claude & Cursor
Drop-in MCP server exposing all six Reducto endpoints — parse, extract, split, edit, classify, upload — to any MCP-compatible agent.
Get the server →