JIT-Agent Paper: Dynamic Harnesses Let Small Models Beat Giants While Cutting Costs by 36%
rohanpaul_ai · x · 2026-09-02
A new paper titled "JIT-Agent: Scaling Harness Intelligence via Just-in-Time Harness Evolution" proposes generating task-specific agent harnesses (memory, planning, tools) on demand, enabling smaller models to outperform stronger general models.
Key Findings:
- Performance Boost: DeepSeek-V4-Flash with JIT-Agent scored 85.1 on DeepSearchQA, beating GPT-5.6 (76.0). Across 18 pairs, JIT improved every backbone, with average gains of 7.7 points for GLM-5.2 and 8.8 for DeepSeek-V4-Flash.
- Cost Reduction: This wasn't just more inference. Compared to fixed harnesses like Claude Code and Codex, JIT-Agent achieved the lowest token usage and API cost in all 6 settings, averaging 36.0% lower cost than the cheapest fixed alternative.
Related event: JIT-Agent Paper Lets Small Models Outperform GPT-5.6 at Lower Cost(2 posts)→
More from coding & agent
- LLM one-shots a Minecraft 3x3 piston door by building its own Linux test infra — Vjeux · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Solo founder builds a money-printing data service on Claude's Pi Durable harness — MicahBerkley · 2026-10-18
- Dev builds Arch-based OS with custom desktop and browser using Claude Code — Snoo-10493 · 2026-10-18
- Five SaaS agent workflows fail the same way — and it's never model intelligence — usehive · 2026-10-18
- Claude skill generates hand-drawn interactive diagrams that teach kids how things work — CatAstro_Piyush · 2026-10-18