FinCrime Agent Benchmark
Agents screen customers for sanctions, political exposure, and adverse media. A case passes when agents find the relevant alerts and resolve them under the bank's policy.
Castellum.AI publishes results for nine models on 500 cases. Each model ran three times with web search and three times with the company's screening system.
- Made by
- Castellum.AI
- Released
- September 2026
- Kind
- Finance
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- Kept private
- Measures
- Capability
- Subject
- Economics and business, Law
- Language
- English
- Interaction
- Agentic
- Environment
- Browser, APIs and tools
- Refresh
- One fixed set
- Run by
- A host
- In views
- Agents, Law and finance
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
Castellum.AI Harness + Claude Sonnet 5 tops the leaderboard with 99.6% GPT-6 Astra leads with 71.1% among models released by September 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model on its own is GPT-6 Astra, at 71.1%. Agents built around a model, such as Castellum.AI Harness + Claude Sonnet 5, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- Claude Opus 5.5
- and 14 more
- 99.6 Castellum.AI Harness + Claude Sonnet 5
- 99.6 Castellum.AI Harness + Gemini 3.6 Flash
- 99.5 Castellum.AI Harness + Claude Fable 5.1
- 99.3 Castellum.AI Harness + GPT-6 Astra
- 99.2 Castellum.AI Harness + Grok 4.20
- 99.2 Castellum.AI Harness + Gemma 4 31B IT
- 99.2 Castellum.AI Harness + GPT-5.5
- 99.2 Castellum.AI Harness + Muse Spark 1.2
- 98.1 Castellum.AI Harness + Kimi K2.5
- 71.1 GPT-6 Astra
- 64.7 GPT-5.5
- 64.3 Claude Fable 5.1
- 60.3
Gemini 3.6 Flash
- 53.6
Gemma 4 31B IT
- 50.6 Kimi K2.5
- 48.2 Claude Sonnet 5
- 45.4 Grok 4.20
- 30.8
Muse Spark 1.2
- 1 Castellum.AI Harness + Claude Sonnet 5 99.6%
- 2 Castellum.AI Harness + Gemini 3.6 Flash 99.6%
- 3 Castellum.AI Harness + Claude Fable 5.1 99.5%
- 4 Castellum.AI Harness + GPT-6 Astra 99.3%
- 5 Castellum.AI Harness + Grok 4.20 99.2%
- 6 Castellum.AI Harness + Gemma 4 31B IT 99.2%
- 7 Castellum.AI Harness + GPT-5.5 99.2%
- 8 Castellum.AI Harness + Muse Spark 1.2 99.2%
- 9 Castellum.AI Harness + Kimi K2.5 98.1%
- 10 GPT-6 Astra 71.1%
- 11 GPT-5.5 64.7%
- 12 Claude Fable 5.1 64.3%
- 13
Gemini 3.6 Flash 60.3%
- 14
Gemma 4 31B IT 53.6%
- 15 Kimi K2.5 50.6%
- 16 Claude Sonnet 5 48.2%
- 17 Grok 4.20 45.4%
- 18
Muse Spark 1.2 30.8%
Source: Official FinCrime Agent Benchmark leaderboard
- 1 GPT-6 Astra 71.1%
- 2 GPT-5.5 64.7%
- 3 Claude Fable 5.1 64.3%
- 4
Gemini 3.6 Flash 60.3%
- 5
Gemma 4 31B IT 53.6%
- 6 Kimi K2.5 50.6%
- 7 Claude Sonnet 5 48.2%
- 8 Grok 4.20 45.4%
- 9
Muse Spark 1.2 30.8%
Source: Official FinCrime Agent Benchmark leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Castellum.AI Harness + Claude Sonnet 5 | 99.6% | Jun 2026 | Official leaderboard | |
| 2 | Castellum.AI Harness + Gemini 3.6 Flash | 99.6% | Jul 2026 | Official leaderboard | |
| 3 | Castellum.AI Harness + Claude Fable 5.1 | 99.5% | Sep 2026 | Official leaderboard | |
| 4 | Castellum.AI Harness + GPT-6 Astra | 99.3% | Sep 2026 | Official leaderboard | |
| 5 | Castellum.AI Harness + Grok 4.20 | 99.2% | Feb 2026 | Official leaderboard | |
| 6 | Castellum.AI Harness + Gemma 4 31B IT | 99.2% | Apr 2026 | Official leaderboard | |
| 7 | Castellum.AI Harness + GPT-5.5 | 99.2% | Apr 2026 | Official leaderboard | |
| 8 | Castellum.AI Harness + Muse Spark 1.2 | 99.2% | Aug 2026 | Official leaderboard | |
| 9 | Castellum.AI Harness + Kimi K2.5 | 98.1% | Jan 2026 | Official leaderboard | |
| 10 | GPT-6 Astra | 71.1% | Standalone | Sep 2026 | Official leaderboard |
| 11 | GPT-5.5 | 64.7% | Standalone | Apr 2026 | Official leaderboard |
| 12 | Claude Fable 5.1 | 64.3% | Standalone | Sep 2026 | Official leaderboard |
| 13 | | 60.3% | Standalone | Jul 2026 | Official leaderboard |
| 14 | | 53.6% | Standalone | Apr 2026 | Official leaderboard |
| 15 | Kimi K2.5 | 50.6% | Standalone | Jan 2026 | Official leaderboard |
| 16 | Claude Sonnet 5 | 48.2% | Standalone | Jun 2026 | Official leaderboard |
| 17 | Grok 4.20 | 45.4% | Standalone | Feb 2026 | Official leaderboard |
| 18 | | 30.8% | Standalone | Aug 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 71.1% | Standalone | Sep 2026 | Official leaderboard |
| 2 | GPT-5.5 | 64.7% | Standalone | Apr 2026 | Official leaderboard |
| 3 | Claude Fable 5.1 | 64.3% | Standalone | Sep 2026 | Official leaderboard |
| 4 | | 60.3% | Standalone | Jul 2026 | Official leaderboard |
| 5 | | 53.6% | Standalone | Apr 2026 | Official leaderboard |
| 6 | Kimi K2.5 | 50.6% | Standalone | Jan 2026 | Official leaderboard |
| 7 | Claude Sonnet 5 | 48.2% | Standalone | Jun 2026 | Official leaderboard |
| 8 | Grok 4.20 | 45.4% | Standalone | Feb 2026 | Official leaderboard |
| 9 | | 30.8% | Standalone | Aug 2026 | Official leaderboard |
Rankings by part
We read 8 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Screening overall, Screening sanctions and Screening PEPs. GPT-6 Astra leads 3 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Screening overall
9 models we track have a score here.
- 1 GPT-6 Astra 81.1%
- 2 Claude Fable 5.1 ↑1 78%
- 3 GPT-5.5 ↓1 75.3%
- 4
Gemini 3.6 Flash 67.3%
- 5 Claude Sonnet 5 ↑2 59.7%
- 6
Gemma 4 31B IT ↓1 58.1%
- 7 Kimi K2.5 ↓1 53.3%
- 8 Grok 4.20 51.4%
- 9
Muse Spark 1.2 40.5%
Screening sanctions
9 models we track have a score here.
- 1 GPT-6 Astra 94.2%
- 2 Claude Fable 5.1 ↑1 87.8%
- 3 Kimi K2.5 ↑3 85.8%
- 4
Gemma 4 31B IT ↑1 85.8%
- 5 GPT-5.5 ↓3 85.8%
- 6
Gemini 3.6 Flash ↓2 74.2%
- 7 Grok 4.20 ↑1 73.3%
- 8 Claude Sonnet 5 ↓1 66.7%
- 9
Muse Spark 1.2 35.8%
Screening PEPs
9 models we track have a score here.
- 1 GPT-6 Astra 92.3%
- 2 Claude Fable 5.1 ↑1 87.1%
- 3 GPT-5.5 ↓1 86.7%
- 4
Gemini 3.6 Flash 86.7%
- 5 Claude Sonnet 5 ↑2 79%
- 6
Muse Spark 1.2 ↑3 75.4%
- 7
Gemma 4 31B IT ↓2 75%
- 8 Kimi K2.5 ↓2 71.2%
- 9 Grok 4.20 ↓1 49.8%
Screening adverse media
9 models we track have a score here.
- 1 Claude Fable 5.1 ↑2 64%
- 2 GPT-6 Astra ↓1 56%
- 3 GPT-5.5 ↓1 54.8%
- 4
Gemini 3.6 Flash 42.6%
- 5 Grok 4.20 ↑3 38.6%
- 6 Claude Sonnet 5 ↑1 37.9%
- 7
Gemma 4 31B IT ↓2 22.6%
- 8 Kimi K2.5 ↓2 11.2%
- 9
Muse Spark 1.2 10.5%
Adjudication overall
9 models we track have a score here.
- 1 Kimi K2.5 ↑5 94.9%
- 2
Gemma 4 31B IT ↑3 92.3%
- 3
Gemini 3.6 Flash ↑1 89.5%
- 4 Grok 4.20 ↑4 88.3%
- 5 GPT-6 Astra ↓4 87.7%
- 6 GPT-5.5 ↓4 85.9%
- 7 Claude Fable 5.1 ↓4 82.6%
- 8 Claude Sonnet 5 ↓1 80.7%
- 9
Muse Spark 1.2 76.1%
Adjudication sanctions
9 models we track have a score here.
- 1
Gemini 3.6 Flash ↑3 97.4%
- 2 GPT-6 Astra ↓1 97.3%
- 3 Kimi K2.5 ↑3 95.5%
- 4
Gemma 4 31B IT ↑1 94.2%
- 5 GPT-5.5 ↓3 92.2%
- 6 Claude Fable 5.1 ↓3 89.5%
- 7 Claude Sonnet 5 82.7%
- 8 Grok 4.20 81.1%
- 9
Muse Spark 1.2 62.8%
Adjudication PEPs
9 models we track have a score here.
- 1 Kimi K2.5 ↑5 98.5%
- 2 GPT-6 Astra ↓1 97.7%
- 3 GPT-5.5 ↓1 97.6%
- 4
Gemini 3.6 Flash 97.4%
- 5 Claude Fable 5.1 ↓2 96.6%
- 6 Grok 4.20 ↑2 96.2%
- 7
Gemma 4 31B IT ↓2 95.6%
- 8 Claude Sonnet 5 ↓1 93.4%
- 9
Muse Spark 1.2 84%
Adjudication adverse media
9 models we track have a score here.
- 1 Grok 4.20 ↑7 89.5%
- 2
Gemma 4 31B IT ↑3 73.7%
- 3
Gemini 3.6 Flash ↑1 66.5%
- 4 Kimi K2.5 ↑2 63.8%
- 5 GPT-5.5 ↓3 53.9%
- 6 GPT-6 Astra ↓5 51.9%
- 7 Claude Fable 5.1 ↓4 50.9%
- 8
Muse Spark 1.2 ↑1 50%
- 9 Claude Sonnet 5 ↓2 46.5%
2 models have held number one on the official leaderboard since its release in September 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official FinCrime Agent Benchmark leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official FinCrime Agent Benchmark leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access The tasks are private.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (18 entries).
- Pass Models scored 9 models we track, plus 9 other entries on its leaderboard.
- Pass Newest score The newest is from September 2026, for GPT-6 Astra.
- Note Frontier models Only 2 of the top 10 have a score. The strongest tested is GPT-6 Astra (#4).
- Pass Headroom The best is 71.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on castellum.ai: 18 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.