MOUNTAIN VIEW, Calif., July 29, 2026 /PRNewswire/ — Samaya AI, the leading AI platform for investment professionals, today published evaluation results for leading frontier models, open-source models, and Samaya’s proprietary AI system on FrontierFinance, the benchmark Samaya previously released for measuring AI Agents on challenging agentic workflows that span the full investment process. Samaya’s AI system outperformed every model evaluated.

System performance

FrontierFinance is the largest and most challenging open benchmark for measuring performance of AI Agents across the full investment workflow. Built by finance experts, it evaluates AI capabilities across all stages of the investment process, from idea screening and research to financial modeling, portfolio tracking, and catalyst monitoring. It consists of 220 diverse queries and 11,543 expert-crafted rubrics, built through Samaya AI’s Criteria Eval framework, for comprehensive grading of Agent outputs.

FrontierFinance is substantially harder than existing finance benchmarks. Existing benchmarks focus almost exclusively on financial data extraction, while FrontierFinance covers a diverse range of use cases that are (i) central to the investment process, (ii) much more difficult to complete, and (iii) much harder to collect high quality evaluation data for. The use cases include Screening & Discovery, Company Research, Sector, Industry & Macro Analysis, Earnings and Events, and Coverage & Catalyst Monitoring. The full FrontierFinance benchmark is available at https://research.samaya.ai/benchmarks/frontier-finance.

Multiple frontier models were evaluated on FrontierFinance, including Anthropic Claude Fable 5 and Claude Opus 4.8, OpenAI GPT-5.6 Sol, and Google Gemini 3.1 Pro, alongside open-source models including GLM 5.2 and DeepSeek V4 Pro, and Samaya’s proprietary AI system. This evaluation was done using the fixed, publicly released rubric set and a standard harness for financial tasks. Samaya’s AI system outperformed all models, with Samaya (low effort) scoring 50.8% accuracy at 4x lower inference cost than Fable 5, and Samaya (high effort) scoring 56% accuracy at 2x lower cost than Fable 5. Anthropic’s Claude Fable 5 was the best performing frontier model, scoring 49.2%, followed by GPT-5.6 Sol at 46.8% and Claude Opus 4.8 at 45%. The full results are available at https://research.samaya.ai/benchmarks/frontier-finance/system-performance.

“Investment use cases are uniquely hard for AI because being almost correct is still a loss. Producing an expert-level response requires accuracy on every datapoint and every step of reasoning across a long, complex workflow, not just a plausible looking output,” said Maithra Raghu, CEO and Founder of Samaya AI. “FrontierFinance is the first benchmark that measures Agents at that fidelity across the investment process. It came out of the work we’ve been doing building at the frontier of AI for investing, and it’s significant to see Samaya’s AI system outperform the best frontier models at a fraction of the cost.”

The FrontierFinance benchmark, evaluation code and results are publicly available at https://research.samaya.ai/benchmarks/frontier-finance. Samaya AI created FrontierFinance from a large internal evaluation set of 5000 complex examples, created over multiple years of designing frontier Finance AI Agents. Samaya plans to release subsequent, harder evaluation benchmarks in future versions.

About Samaya AI

Samaya AI builds Expert AI Agents for investment professionals, supporting high-stakes investment workflows across leading financial institutions. By combining finance-specific AI models with deep reasoning over institutional context and global markets, Samaya helps investors go from information to conviction. Samaya is deployed at enterprise scale across hedge funds, asset managers, and banks, and serves tens of thousands of investors globally. Backed by leading AI investors including NVIDIA, NEA, Databricks, Eric Schmidt, and Yann LeCun, Samaya continues to advance the next generation of AI for investment professionals.

Compare systems