Skip to main content
AI Fin Hub
Menu

Benchmarks

Dated reports on the costs and tools of AI in markets, each reproducible: the method is published first, the sources are linked where they are used, and the data lives in the tools that produced it.

Releases 3

  • State of LLM pricing for finance, Q3 2026

    Pricing-grade snapshot of every finance-relevant LLM tier as of 2026-06-21, sourced line-by-line from vendor pricing pages. Six providers: Anthropic (Opus 4.8, Fable 5, Sonnet 4.6, Haiku 4.5), OpenAI GPT-5.5, Google Gemini 3.5 Flash + 2.5, xAI Grok 4.3, DeepSeek V4, Mistral — with four worked finance scenarios. Supersedes Q2.

    Q3 release, released . 21 model tiers across 6 vendors. 4 worked finance scenarios with $/month. Cache, batch, and context-window rates documented. Downloadable CSV + JSON.

  • State of LLM pricing for finance, Q2 2026

    Pricing-grade snapshot of every finance-relevant LLM tier as of 2026-04-23, sourced line-by-line from vendor pricing pages. Anthropic Claude Haiku 4.5 / Sonnet 4.6 / Opus 4.6 + 4.7, OpenAI GPT-5.4 family, Google Gemini 2.5 — with four worked finance scenarios. Superseded by the Q3 2026 release.

    Q2 release (superseded by Q3), released . 14 model tiers across 3 vendors. 4 worked finance scenarios with $/month. Cache, batch, context, and thinking rates documented. Downloadable CSV + JSON.

  • State of AI market data 2026

    The 2026 baseline for how retail and small-team AI operators source, use, and pay for market data across Databento, Massive (formerly Polygon.io), Alpaca, Tiingo, FMP, and Alpha Vantage — plus MCP coverage by vendor and LLM-task price/performance.

    Q2 release, released . 6 vendors benchmarked. 6 LLM models cost-modeled on a finance research loop. MCP coverage audit. Reproducible via linked tools.

Every benchmark follows the same rules

Methodology first
Every benchmark publishes its full methodology before the report.
Reproducible figures
The LLM pricing reports ship CSV + JSON downloads; every report documents how to reproduce its figures.
Editorially independent
No sponsors, affiliate links, or paid placements in any benchmark.
Public corrections
All corrections logged at /corrections/ with dated entries.
Frozen at publication
Reports are dated at publication. Revisions ship as a new report.