Key Takeaways
- • Samaya AI released FrontierFinance, the largest and most challenging public benchmark for evaluating AI agents across full investment workflows, comprising 220 diverse queries and 11,543 expert-crafted rubrics built through its Criteria Eval framework.
- • Samaya's proprietary AI system achieved state-of-the-art performance at 50.8% accuracy on FrontierFinance, outperforming Anthropic's Claude Fable 5 (49.2%), while operating at approximately 4x lower inference cost than Claude Fable 5.
- • FrontierFinance benchmark covers critical investment process stages including idea screening, company research, sector and macro analysis, earnings analysis, and catalyst monitoring—substantially more comprehensive than existing finance benchmarks that focus primarily on data extraction.
- • Multiple frontier models were evaluated including OpenAI's GPT-5.6 Sol, Google Gemini 3.1 Pro, Claude Opus 4.8, and open-source models GLM 5.2 and DeepSeek V4 Pro, with Samaya's system outperforming all competitors across the evaluation.
- • Independent testing confirms Samaya's system delivers 96% accuracy compared to 57% for leading GPT-based systems, positioning the benchmark as a critical tool for measuring AI agent performance in high-stakes financial workflows.
Summary
Samaya AI released evaluation results demonstrating its proprietary AI system's superior performance on FrontierFinance, a newly published open benchmark designed to measure AI agent capabilities across the complete investment workflow. The benchmark represents a significant advancement in evaluating frontier intelligence for financial applications, addressing a critical gap in existing evaluation methodologies that focus narrowly on data extraction rather than comprehensive agentic workflows.
FrontierFinance consists of 220 diverse queries evaluated against 11,543 expert-crafted rubrics developed through Samaya's proprietary Criteria Eval framework. The benchmark encompasses five primary use cases central to institutional investment processes: screening and discovery, company research, sector and macro analysis, earnings and events analysis, and coverage and catalyst monitoring. This comprehensive scope substantially exceeds existing finance benchmarks in both breadth and difficulty, requiring AI systems to demonstrate sophisticated reasoning across multiple stages of the investment decision-making process rather than simple financial data retrieval.
Samaya's in-house AI system achieved state-of-the-art performance at 50.8% accuracy on FrontierFinance, outperforming Anthropic's Claude Fable 5 at 49.2% accuracy while maintaining approximately 4x lower inference costs. Additional models evaluated included Claude Opus 4.8 (45%), OpenAI's GPT-5.6 Sol, Google Gemini 3.1 Pro, and open-source competitors GLM 5.2 and DeepSeek V4 Pro. Independent testing results indicated Samaya's system achieved 96% accuracy relative to 57% for leading GPT-based alternatives, underscoring the performance differential in production environments.
The timing of FrontierFinance's release coincides with OpenAI's announcement of GPT-5.6 Sol, positioning Samaya's benchmark as a timely evaluation tool for institutional investors and asset managers assessing AI agent capabilities for investment workflows. The benchmark is publicly available and represents Samaya's commitment to transparent, expert-driven evaluation methodologies for AI systems operating in high-stakes financial environments where accuracy, cost efficiency, and comprehensive reasoning across multi-stage workflows are essential for institutional adoption.
FrontierFinance consists of 220 diverse queries evaluated against 11,543 expert-crafted rubrics developed through Samaya's proprietary Criteria Eval framework. The benchmark encompasses five primary use cases central to institutional investment processes: screening and discovery, company research, sector and macro analysis, earnings and events analysis, and coverage and catalyst monitoring. This comprehensive scope substantially exceeds existing finance benchmarks in both breadth and difficulty, requiring AI systems to demonstrate sophisticated reasoning across multiple stages of the investment decision-making process rather than simple financial data retrieval.
Samaya's in-house AI system achieved state-of-the-art performance at 50.8% accuracy on FrontierFinance, outperforming Anthropic's Claude Fable 5 at 49.2% accuracy while maintaining approximately 4x lower inference costs. Additional models evaluated included Claude Opus 4.8 (45%), OpenAI's GPT-5.6 Sol, Google Gemini 3.1 Pro, and open-source competitors GLM 5.2 and DeepSeek V4 Pro. Independent testing results indicated Samaya's system achieved 96% accuracy relative to 57% for leading GPT-based alternatives, underscoring the performance differential in production environments.
The timing of FrontierFinance's release coincides with OpenAI's announcement of GPT-5.6 Sol, positioning Samaya's benchmark as a timely evaluation tool for institutional investors and asset managers assessing AI agent capabilities for investment workflows. The benchmark is publicly available and represents Samaya's commitment to transparent, expert-driven evaluation methodologies for AI systems operating in high-stakes financial environments where accuracy, cost efficiency, and comprehensive reasoning across multi-stage workflows are essential for institutional adoption.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.