Dalpha Tops 3 Global AI Benchmarks, Beating Google and OpenAI

Takes first place across three major technical benchmarks, from time-series forecasting to deep research

AI agent company Dalpha announced on June 22 that it had once again demonstrated world-class technology, ranking first across three global AI benchmarks and beating leading global big tech companies.

The achievement comes from a head-to-head contest with major global IT and AI giants such as Google, OpenAI, and NVIDIA — companies backed by vast capital, talent, and computing scale. What makes it especially significant is that Dalpha delivered top performance across domains not through a single model specialized for one task, but through a general-purpose agent framework the company developed in-house.

The three benchmarks Dalpha topped are core global metrics spanning time-series forecasting and deep research.

First, Salesforce's "GIFT-Eval" is a general-purpose time-series forecasting benchmark comprising 28 datasets and more than 140,000 time series. Dalpha's agentic system outperformed foundation models trained specifically for time series by the likes of Datadog and Salesforce, taking the overall top spot by posting the most consistently high rankings across a wide range of datasets.

Time-series forecasting — the technology of analyzing data accumulated over time to predict the future — is regarded as a core capability directly tied to product demand forecasting, logistics optimization, and marketing for consumer packaged goods (CPG) brands in fashion, beauty, F&B, and beyond.

On "DeepResearch Bench," which tackles 100 PhD-level research tasks across 22 domains, Dalpha also ranked first overall, surpassing the flagship deep research agents of big tech players including Google Gemini, OpenAI, Perplexity, and xAI's Grok.

Dalpha also swept "DeepResearch Bench II (rigorous evaluation)," considered one of the most stringent agent benchmarks. Evaluated against 9,430 fine-grained rubrics derived from 132 expert reports, Dalpha outscored OpenAI, Google, NVIDIA, Huawei, Alibaba, and ByteDance on the overall metric.

Earlier, in May, Dalpha had already recorded a high score of 79.11% on "MLE-bench," OpenAI's official agent performance benchmark, surpassing Google and Baidu in its in-house technical validation.

A lean, elite team going head-to-head with global big tech — proving the technological strength of "K-AI"

These results are drawing industry attention precisely because they came not from a large corporate organization but from a small, elite team made up of some of Korea's top talent. Dalpha's co-founders are graduates of Seoul National University and Seoul Science High School, and its AI team includes numerous Mathematical Olympiad medalists. It is a case proving that "talent density," rather than sheer capital scale, can secure a competitive edge against global big tech.

Dogyun Kim, CEO of Dalpha, said, "In a global AI race led by enormous capital and large organizations, what I'm most proud of is that Dalpha's small, elite team swept three benchmarks across entirely different domains at once and beat the big tech players across the board." He added, "Together with this exceptional team, Dalpha is now building a next-generation foundation model that understands human behavior — what we call a 'Social World Model.' As we always have, we will be the first to prove out the next wave of the AI market, built on our technical achievements."