2026-09-10MODELDeepSeek-V4.1-Flash released with extreme agent memory efficiency and evaluated across VALS • 2026-09-03MODELGPT-6 Astra released by OpenAI - 71.7% on EMB, 63.3% on Tax Agent Bench • 2026-09-02MODELGoogle Gemini 3.8 Flash takes #1 rank on Finance Agent (v2) with 61.4% and 72.2% on EMB • 2026-09-02MODELMeta releases Muse Spark 1.3 & 1.3 Max - 59.0%-60.0% on Finance Agent (v2) • 2026-09-01MODELClaude Fable 5.1 leads Tax Agent Bench (77.6%) and EMB (76.7%) with 75% cheaper cache reads • 2026-09-11BENCHTax Agent Bench and FAB v2 benchmarks updated with latest frontier model rankings • 2026-09-15UPDATEPlatform updated: 43 models tracked across 41 independent benchmarks • 2026-09-10MODELDeepSeek-V4.1-Flash released with extreme agent memory efficiency and evaluated across VALS • 2026-09-03MODELGPT-6 Astra released by OpenAI - 71.7% on EMB, 63.3% on Tax Agent Bench • 2026-09-02MODELGoogle Gemini 3.8 Flash takes #1 rank on Finance Agent (v2) with 61.4% and 72.2% on EMB • 2026-09-02MODELMeta releases Muse Spark 1.3 & 1.3 Max - 59.0%-60.0% on Finance Agent (v2) • 2026-09-01MODELClaude Fable 5.1 leads Tax Agent Bench (77.6%) and EMB (76.7%) with 75% cheaper cache reads • 2026-09-11BENCHTax Agent Bench and FAB v2 benchmarks updated with latest frontier model rankings • 2026-09-15UPDATEPlatform updated: 43 models tracked across 41 independent benchmarks •
Home/Research
FinanceBenchmark Research

Research

Analysis, methodology notes, and findings from our evaluation work. Written for practitioners building or deploying AI in financial services.

Filter
Latest
State of Financial AIJun 11, 20269 min read
The state of financial AI benchmarking in 2026
Frontier models have reached human-level performance on simple financial QA. But verification - knowing whether an answer is right rather than just plausible - remains largely unsolved. We look at where the field stands and what it will take to close the gap.
Read article →
Model EvaluationJun 18, 2026
Why forensic discovery tasks reveal what frontier models still cannot do
Document QA benchmarks make frontier models look better than they are. Forensic discovery - finding what is missing, inconsistent, or deliberately obscured - is a different task. We explain why, and what happens when you test for it.
7 min readRead →
02
MethodologyJun 25, 2026
How the FB Composite Score is calculated
A plain-language explanation of the weighted average behind the FB Score - what the tier weights mean, why we normalise by evaluated benchmarks rather than total benchmarks, and what the asterisk means.
5 min readRead →
03
About this publication

FinanceBenchmark Research publishes analysis on AI evaluation in financial services — what the numbers mean, where models fail, and what good financial AI reasoning actually looks like.

Our methodology →All benchmarks →
Recent posts
State of Financial AIJun 11, 2026
The state of financial AI benchmarking in 2026
Model EvaluationJun 18, 2026
Why forensic discovery tasks reveal what frontier models still cannot do
MethodologyJun 25, 2026
How the FB Composite Score is calculated
Topics