Driven Launches DrivenBench to Compare Leading AI Models Across Real-World Investment Tasks

New benchmark evaluates 11 frontier models across capability, cost and latency to help investors match models to finance-specific workflows.

Singapore, Singapore, September 25, 2026 -- SINGAPORE, September 29, 2026 - Driven, the AI-powered investment agent platform, has announced DrivenBench 1.0, a benchmark designed to compare leading AI models on the real-world investment tasks agents are expected to perform. In the DrivenBench 1.0 capability ranking, Anthropic's Claude Sonnet 5 and Moonshot AI's Kimi K3 tied for first place with scores of 93.9%. Claude Opus 5 ranked third at 90.9%, followed by DeepSeek V4 Pro at 89.7% and xAI's Grok 4.6 at 89.1%. The full leaderboard also publishes cost estimates and median and p90 latency so users can assess performance alongside efficiency and responsiveness.

Rather than relying on broad, general-purpose tests, DrivenBench evaluates models inside Driven using finance-specific workflows and reports results across capability, cost and latency.

The first release compares 11 models through 55 formal capability evaluations grounded in 15 investment task categories and a test set of more than 160 questions. The benchmark is intended to show where each model performs well, where it may fail and what trade-offs investors face when choosing a model for research, analysis and monitoring.

Why finance-specific model evaluation matters

Investment workflows place demands on AI systems that generic benchmarks do not fully capture. A model may need to retrieve current market information, interpret filings and financial statements, compare valuation assumptions, identify catalysts, maintain portfolio context and produce an output that can be checked against its sources. Speed and cost also matter when agents run recurring tasks or monitor markets throughout the day.

DrivenBench measures model performance in this operating context. Its capability score is calculated as the equally weighted average pass rate across the evaluations. Separate regression checks cover critical failure modes without affecting the ranking, while cost figures use published-price upper bounds and latency figures provide directional comparisons based on observed runtimes.

The benchmark supports Driven's broader approach to building more reliable investment agents. Driven gives users access to multiple frontier model families and offers both manual selection and an Auto mode. By continuously testing models as they change, Driven can better match different models to different investment tasks and improve the reliability, efficiency and task awareness of its Agents.

“Investment Agents need more than access to powerful models. Users also need a reliable way to identify which model is best suited to each financial task,” said Zania Zhang, Community Manager at Driven. “The real test is whether an Agent remains dependable in complex, real-world investment workflows — particularly in areas such as options analysis, where data quality, calculations and strategy trade-offs all matter. DrivenBench reflects our broader commitment to building a more reliable AI investment agent platform, helping users move from scattered information to better-supported investment decisions.”

Built for agents that work with real investment context

Driven is designed to give each investor an AI investment team rather than a general-purpose chatbot. The platform connects its Agents to more than 260 investment APIs spanning real-time quotes, fundamentals, filings, ownership, insider activity, earnings, macroeconomic data, options, news, technical indicators and portfolio context.
Investment information is abundant, but turning it into timely, decision-ready insight remains difficult. Options research provides a clear example of why this matters. Analyzing an options position can require real-time option-chain data, implied volatility, Greeks, volume and open interest, together with portfolio exposure and trade-offs involving timing, risk and strategy. Driven brings these inputs into one Agent workspace, combining real-time U.S. and Hong Kong stock and ETF options data with dedicated options tools, investment Skills, portfolio analysis, and supported broker connections. This helps users move from contract screening and scenario analysis to portfolio-aware research and, where available, supervised execution. Its investment skills turn repeatable research methods into structured processes for tasks such as valuation analysis, stock screening, macro analysis, earnings review, IPO analysis and portfolio monitoring. Through the Driven Skill Store, users can discover reviewed and versioned Skills, install them in their Agents and publish their own methods for other investors to use, making professional research workflows easier to reuse and share.

Driven Connectors extend that workflow beyond the chat window. Supported brokerage accounts, professional data providers and productivity tools can give an Agent access to real portfolio context, subscribed datasets and existing research files. Users can also create Playbooks that capture an investment thesis, selection criteria, risk preferences and position-sizing rules, then apply those Playbooks across portfolios and scheduled monitoring tasks.

Paper trading is currently available, allowing investors to test and refine strategies before acting. Broker-connected order capabilities depend on the supported connector and account configuration, while investment decisions remain under the user's control.

Built for investors, adopted by investors

Since launch, Driven has attracted a growing community of investors and finance professionals. The platform is already trusted by portfolio managers overseeing more than $500 million in assets, as well as engineers working on frontier AI. Users are applying Driven across recurring market monitoring, portfolio review, earnings analysis, options research and other real-world investment workflows.

This adoption reflects Driven's broader mission: to make professional-grade investment research more accessible, reliable and repeatable. Driven is building toward a more dependable AI investment team that helps investors research, monitor and act with greater confidence.

View the complete DrivenBench methodology and leaderboard at driven.ai/drivenbench.

About Driven

Driven is an AI-powered investment agent platform developed by SNOWBALL WEALTH PRIVATE LIMITED. Built by investors and builders, Driven gives every investor an AI investment team powered by real market data, 260+ investment APIs, reusable investment Skills, a dedicated Options Space, scheduled monitoring, evolving Playbooks and portfolios, and broker-connected execution. Driven is available on desktop and mobile web browsers. Your Agents handle the work. You make the calls.

Disclaimer: DrivenBench results are comparative research based on the tested model versions and conditions and may change over time. Driven does not provide investment advice. AI-generated analysis may contain errors, and users remain responsible for verifying information and making investment decisions.

Contact Info:
Name: Zania Zhang
Email: Send Email
Organization: Driven (operated by SNOWBALL WEALTH PRIVATE LIMITED)
Phone: 15550100
Website: https://driven.ai/

Release ID: 89204395

Should you identify any discrepancies, concerns, or inaccuracies in the content provided in this press release or require assistance with a press release takedown, we strongly urge you to notify us promptly by contacting [email protected] (it is important to note that this email is the authorized channel for such matters, sending multiple emails to multiple addresses does not necessarily help expedite your request). Our responsive team is committed to addressing your concerns within 8 hours by taking necessary actions to resolve identified issues diligently or guiding you through the necessary steps for removal. Our dedication lies in providing accurate and reliable information.