Jan-nano Technical Report

TL;DR

Jan-nano is a highly efficient 4B parameter language model that excels in instant retrieval and reasoning tasks, achieved through novel training methods and specialization, enabling high performance on consumer hardware.

Contribution

The paper introduces Jan-nano, a specialized 4B model trained with a new RLVR system that eliminates traditional next token prediction, demonstrating high efficiency and performance.

Findings

Achieves 83.2% on SimpleQA benchmark

Operates effectively with 128K context length

Runs on consumer hardware with high efficiency

Abstract

Most language models face a fundamental tradeoff where powerful capabilities require substantial computational resources. We shatter this constraint with Jan-nano, a 4B parameter language model that redefines efficiency through radical specialization: instead of trying to know everything, it masters the art of finding anything instantly. Fine-tuned from Qwen3-4B using our novel multi-stage Reinforcement Learning with Verifiable Rewards (RLVR) system that completely eliminates reliance on next token prediction training (SFT), Jan-nano achieves 83.2% on SimpleQA benchmark with MCP integration while running on consumer hardware. With 128K context length, Jan-nano proves that intelligence isn't about scale, it's about strategy.