OpenVals

OpenVals Index

UPDATED 10/8/2026|
VERSION 0.5.5

A single measure of AI's potential economic impact — agentic model performance across finance, coding, and legal tasks, weighted by each sector's share of U.S. GDP.

OpenVals IndexGDP-weighted benchmark
SHOWING LATEST, TOP & FRONTIER MODELS (25)
EFFICIENT (FAST & ACCURATE)INEFFICIENT (SLOW & LOWER ACCURACY)FASTEST (LOW LATENCY)BEST ACCURACYOPENVALS0%20%40%60%80%100%0s10s20s30s40s50sDeepScaler 1.5B (100%)Hunyuan 1.8BGranite 3.3LFM 2.5LLaMA 2.0Phi-3 3.8BOpenChat 7BQwen 3Yi-Coder 1.5BOneLLM DoeyBloom 560MSupra 50MWillow Alpha

Key Takeaways

The Latency Frontier Curve

Non-dominated champions trace the optimal trade-off boundary between latency and accuracy: Willow Alpha (303ms, 12.4%), Supra 50M (581ms, 37.3%), OneLLM Doey (972ms, 47.2%), Yi-Coder (2.1s, 69.8%), Granite 3.3 (2.4s, 96.7%), Hunyuan 1.8B (15.3s, 98.3%), and DeepScaler 1.5B (19.4s, 100.0%).

Decision Reliability (DRS) Leaders

OpenChat 7B (63.1%), LFM 2.5 (61.6%), and Granite 3.3 (61.4%) demonstrate top-tier decision reliability scores, proving exceptional consistency and reasoning stability under complex multi-step agent workflows.

Frontier Ceiling: DeepScaler 1.5B

Agentica's DeepScaler 1.5B commands the non-dominated performance ceiling with 100.0% benchmark score across all evaluated workflows, combining perfect task execution with fast sub-20s response times and strong 57.7% DRS score.

Ultra-Fast Sub-Second Inference

Sub-1B parameter models like Willow Alpha (303ms) and Supra 50M (581ms) prove that high-throughput triage, routing, and data extraction can be served in mere hundreds of milliseconds.