Sahara AI Cites Growing Role in Microsoft’s MATHVISTA Benchmark and AI Data Services

According to a recent LinkedIn post from Sahara AI, the company collaborated with Microsoft Research on MATHVISTA, a benchmark designed to evaluate mathematical reasoning in visual contexts such as images, charts, and diagrams. The post indicates that leading multimodal AI models, including GPT-4V, Gemini, Claude, and Bard, significantly underperform human-level accuracy on these tasks, with GPT-4V reportedly scoring 49.9%, or 10.4 percentage points below human performance.

The post suggests that MATHVISTA is gaining traction as a reference standard, citing more than 275,000 total downloads and 13,000 downloads in the past month, and notes that it is being used by researchers and labs globally. This uptake could strengthen Sahara AI’s positioning as an infrastructure and data partner to major AI developers, potentially increasing its relevance in the broader AI tooling and evaluation ecosystem.

Sahara AI’s role appears focused on high-precision data labeling, with the post stating that Microsoft ran a competitive pilot in which many providers and crowdsourcing platforms reportedly failed to meet the task complexity requirements. According to the post, Sahara AI prevailed by leveraging its Data Services Platform, which it describes as a network of over 200,000 pre-vetted annotators across 35+ countries and 45+ languages, supported by custom training modules and multi-phase quality assurance.

The company highlights that MATHVISTA represents the first phase of an ongoing partnership with Microsoft Research and references relationships with other large technology firms such as Amazon and Snap Inc. For investors, the post points to growing demand for specialized, high-quality annotation and evaluation services in advanced AI workflows, which could translate into recurring enterprise engagements and a stronger competitive moat in the data services segment of the AI value chain.

More broadly, the benchmark results underscored in the post emphasize a measurable gap between current multimodal AI performance and human reasoning in mathematically intensive visual tasks. This gap may signal a sustained need for iterative benchmarking, richer datasets, and continuous model improvement, creating a multi-year runway for companies like Sahara AI that supply the data infrastructure underpinning next-generation AI systems.