# FinanceBenchmark.ai > FinanceBenchmark provides independent rankings and composite evaluations of leading Artificial Intelligence models on complex financial reasoning and forensic accounting benchmarks. This file serves as a structured entry point for Large Language Models and AI agents requesting markdown representations of FinanceBenchmark. All URLs ending in .md return clean, unadorned Markdown with tables and metadata. ## Table of Contents - [Core Resources](#core-resources) - [Top Ranked Models](#top-ranked-models) - [Key Financial AI Benchmarks](#key-benchmarks) - [Research & Analysis](#research) - [Full Dataset & Context](#full-dataset) ## Core Resources - [Leaderboard Home](https://financebenchmark.ai/index.md): Real-time composite rankings across all financial models and benchmarks. - [Evaluation Methodology](https://financebenchmark.ai/methodology.md): Mathematical formula, tier weighting (Tier 1 weight 30, Tier 2 weight 20, Tier 3 weight 10), and attribution principles. - [All Models Directory](https://financebenchmark.ai/models.md): Complete list of 52 frontier, specialized, and open-weight models. - [All Benchmarks Catalog](https://financebenchmark.ai/benchmarks.md): Complete index of 42 financial benchmarks spanning verification, document QA, and agent workflows. - [About FinanceBenchmark](https://financebenchmark.ai/about.md): Institutional mission, governance, and Avae founding disclosure. - [Submit Score](https://financebenchmark.ai/submit.md): Guidelines for peer-reviewed evaluation submissions. - [Model Comparison Tool](https://financebenchmark.ai/comparison.md): Side-by-side domain breakdown comparison. ## Top Ranked Models Composite FB Scores as of 2026-10-20: - [Avae 2.0](https://financebenchmark.ai/models/avae-2-0.md): Composite FB Score **97.0** | Avae | 2 benchmarks evaluated. - [Claude Opus 4.7](https://financebenchmark.ai/models/claude-opus-4-7.md): Composite FB Score **73.8** | Anthropic | 10 benchmarks evaluated. - [GPT-5.5](https://financebenchmark.ai/models/gpt-5-5.md): Composite FB Score **72.2** | OpenAI | 12 benchmarks evaluated. - [GPT-6 Sol](https://financebenchmark.ai/models/gpt-6-sol.md): Composite FB Score **71.1** | OpenAI | 1 benchmarks evaluated. - [DeepSeek V4 Pro](https://financebenchmark.ai/models/deepseek-v4-pro.md): Composite FB Score **70.1** | DeepSeek (Open-Weight) | 9 benchmarks evaluated. - [Gemini 3.1 Pro](https://financebenchmark.ai/models/gemini-3-1-pro.md): Composite FB Score **70.0** | Google | 9 benchmarks evaluated. - [GPT-6 Luna](https://financebenchmark.ai/models/gpt-6-luna.md): Composite FB Score **70.0** | OpenAI | 1 benchmarks evaluated. - [Claude Fable 5](https://financebenchmark.ai/models/claude-fable-5.md): Composite FB Score **69.8** | Anthropic | 25 benchmarks evaluated. - [Kimi K2.6](https://financebenchmark.ai/models/kimi-k2-6.md): Composite FB Score **67.4** | Moonshot AI (Open-Weight) | 9 benchmarks evaluated. - [Claude Opus 5](https://financebenchmark.ai/models/claude-opus-5.md): Composite FB Score **66.7** | Anthropic | 30 benchmarks evaluated. - [Claude Fable 5.1](https://financebenchmark.ai/models/claude-fable-5-1.md): Composite FB Score **66.2** | Anthropic | 5 benchmarks evaluated. - [GLM-5.1](https://financebenchmark.ai/models/glm-5-1.md): Composite FB Score **65.9** | Zhipu AI (Open-Weight) | 7 benchmarks evaluated. - [Grok 4.6](https://financebenchmark.ai/models/grok-4-6.md): Composite FB Score **65.4** | xAI | 21 benchmarks evaluated. - [GPT-5.6 Sol](https://financebenchmark.ai/models/gpt-5-6-sol.md): Composite FB Score **65.1** | OpenAI | 29 benchmarks evaluated. - [Grok 4.3](https://financebenchmark.ai/models/grok-4-3.md): Composite FB Score **65.0** | xAI | 9 benchmarks evaluated. ## Key Benchmarks - [Pave Interchange Fee Benchmark](https://financebenchmark.ai/benchmarks/pave.md): Forensic verification across 50,000 synthetic transaction records spanning 4 complexity levels. - [FinanceBench](https://financebenchmark.ai/benchmarks/financebench.md): Financial question answering over real SEC earnings documents. - [FinQA](https://financebenchmark.ai/benchmarks/finqa.md): Multi-step numerical reasoning over corporate financial disclosures. - [ConvFinQA](https://financebenchmark.ai/benchmarks/convfinqa.md): Multi-turn conversational reasoning over financial tables. - [Finance Agent](https://financebenchmark.ai/benchmarks/finance-agent.md): Autonomous financial research workflows and tool use. - [CorpFin](https://financebenchmark.ai/benchmarks/corpfin.md): Long credit agreement comprehension and covenant compliance. - [FinAuditing](https://financebenchmark.ai/benchmarks/finauditing.md): Multi-document forensic audit discovery and discrepancy tracing. ## Research - [The autumn 2026 frontier wave: Evaluating GPT-6, Claude 5.5, and Gemini 4 on financial reasoning](https://financebenchmark.ai/research/frontier-models-autumn-2026-evaluation.md): The rapid arrival of OpenAI's GPT-6 generation, Anthropic's Claude 5.5 lineup, and Google's Gemini 4 Argon marks a major leap in autonomous agent architectures. Here is how they perform on complex financial verification and forensic accounting. - [The state of financial AI benchmarking in 2026](https://financebenchmark.ai/research/state-of-financial-ai-benchmarking-2026.md): A survey of how financial institutions evaluate AI systems today, and why existing benchmarks leave critical gaps in forensic reasoning and verification. - [Why forensic discovery tasks reveal what frontier models still cannot do](https://financebenchmark.ai/research/why-forensic-discovery-reveals-model-limitations.md): Frontier models excel at question answering but struggle when asked to discover the right questions. Forensic financial tasks expose this gap. - [How the FB Composite Score is calculated](https://financebenchmark.ai/research/how-the-fb-composite-score-is-calculated.md): A transparent explanation of the weighting methodology behind the FB Composite Score, including tier definitions and the rationale for each weight. ## Full Dataset - [Complete Knowledge Base (llms-full.txt)](https://financebenchmark.ai/llms-full.txt): Single-file aggregation of all methodology, benchmarks, models, and scores for full-context ingestion.