MultiFinGround
Models identify accounting concepts and extract every matching fact from English, French, and Greek financial reports. The benchmark checks their selections against annotated disclosures.
- Made by
- University of Manchester and collaborators
- Released
- October 2026
- Kind
- Finance
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Measures
- Capability
- Subject
- Economics and business
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- A host
- In views
- Law and finance
- Strongest model tested
-
Gemini 3.8 Flash #24 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
Gemini 3.8 Flash leads with 95.4% among models released by September 2026 Gemini 3.8 Flash leads with 95.4% among models released by September 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is the first task in Table 3, seven-class macro-F1 for financial fact identification. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. 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 16 more
- 95.4
Gemini 3.8 Flash
- 90.5
Gemma 4 31B IT
- 88.6 GPT-5.6 Luna
- 88.1
DeepSeek V4.1 Flash
- 87.6
Qwen3.5-122B-A10B
- 85.9 Claude Haiku 4.5
- 70.7
Qwen3-8B
- 69.8 ODA-Fin-RL-8B
- 68.1
Qwen 3.8 27B
- 66.6
Qwen3.8-Flash-Next
- 51.0 Fin-o1-8B
- 39.5 Qwen2.5-7B-Instruct
- 37.3 Llama-Krikri-8B-Instruct
- 31.1 Fin-R1
- 12.6 Plutus-8B-Instruct
- 1
Gemini 3.8 Flash Reasoning: off 95.4%
- 2
Gemma 4 31B IT Reasoning: model default (no thinking) 90.5%
- 3 GPT-5.6 Luna Reasoning: off 88.6%
- 4
DeepSeek V4.1 Flash Reasoning: off 88.1%
- 5
Qwen3.5-122B-A10B Reasoning: model default (thinks) 87.6%
- 6 Claude Haiku 4.5 Reasoning: off 85.9%
- 7
Qwen3-8B Reasoning: model default (thinks) 70.7%
- 8 ODA-Fin-RL-8B Reasoning: trained reasoning 69.8%
- 9
Qwen 3.8 27B Reasoning: model default (thinks) 68.1%
- 10
Qwen3.8-Flash-Next Reasoning: model default (thinks) 66.6%
- 11 Fin-o1-8B Reasoning: trained reasoning 51%
- 12 Qwen2.5-7B-Instruct Reasoning: none 39.5%
- 13 Llama-Krikri-8B-Instruct Reasoning: none 37.3%
- 14 Fin-R1 Reasoning: trained reasoning 31.1%
- 15 Plutus-8B-Instruct Reasoning: none 12.6%
Source: The MultiFinGround paper
- 1
Gemini 3.8 Flash Reasoning: off 95.4%
- 2
Gemma 4 31B IT Reasoning: model default (no thinking) 90.5%
- 3 GPT-5.6 Luna Reasoning: off 88.6%
- 4
DeepSeek V4.1 Flash Reasoning: off 88.1%
- 5
Qwen3.5-122B-A10B Reasoning: model default (thinks) 87.6%
- 6 Claude Haiku 4.5 Reasoning: off 85.9%
- 7
Qwen3-8B Reasoning: model default (thinks) 70.7%
- 8
Qwen 3.8 27B Reasoning: model default (thinks) 68.1%
- 9
Qwen3.8-Flash-Next Reasoning: model default (thinks) 66.6%
Source: The MultiFinGround paper
| # | Entry | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 95.4% 94.5 to 96.2 | Reasoning: off | Sep 2026 | Official leaderboard |
| 2 | | 90.5% 89.2 to 91.7 | Reasoning: model default (no thinking) | Apr 2026 | Official leaderboard |
| 3 | GPT-5.6 Luna | 88.6% 87.3 to 89.5 | Reasoning: off | Jul 2026 | Official leaderboard |
| 4 | | 88.1% 87.1 to 88.9 | Reasoning: off | Sep 2026 | Official leaderboard |
| 5 | | 87.6% 86.1 to 88.8 | Reasoning: model default (thinks) | Feb 2026 | Official leaderboard |
| 6 | Claude Haiku 4.5 | 85.9% 84.4 to 87 | Reasoning: off | Oct 2025 | Official leaderboard |
| 7 | | 70.7% 67.9 to 73.3 | Reasoning: model default (thinks) | Apr 2025 | Official leaderboard |
| 8 | ODA-Fin-RL-8B | 69.8% 67.1 to 72.2 | Reasoning: trained reasoning | Official leaderboard | |
| 9 | | 68.1% 65.1 to 70.9 | Reasoning: model default (thinks) | Aug 2026 | Official leaderboard |
| 10 | | 66.6% ±2.9 | Reasoning: model default (thinks) | Aug 2026 | Official leaderboard |
| 11 | Fin-o1-8B | 51% ±2.6 | Reasoning: trained reasoning | Official leaderboard | |
| 12 | Qwen2.5-7B-Instruct | 39.5% 37.2 to 41.7 | Reasoning: none | Official leaderboard | |
| 13 | Llama-Krikri-8B-Instruct | 37.3% ±1.5 | Reasoning: none | Official leaderboard | |
| 14 | Fin-R1 | 31.1% 29 to 33.1 | Reasoning: trained reasoning | Official leaderboard | |
| 15 | Plutus-8B-Instruct | 12.6% 10.5 to 14.8 | Reasoning: none | Official leaderboard |
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 95.4% 94.5 to 96.2 | Reasoning: off | Sep 2026 | Official leaderboard |
| 2 | | 90.5% 89.2 to 91.7 | Reasoning: model default (no thinking) | Apr 2026 | Official leaderboard |
| 3 | GPT-5.6 Luna | 88.6% 87.3 to 89.5 | Reasoning: off | Jul 2026 | Official leaderboard |
| 4 | | 88.1% 87.1 to 88.9 | Reasoning: off | Sep 2026 | Official leaderboard |
| 5 | | 87.6% 86.1 to 88.8 | Reasoning: model default (thinks) | Feb 2026 | Official leaderboard |
| 6 | Claude Haiku 4.5 | 85.9% 84.4 to 87 | Reasoning: off | Oct 2025 | Official leaderboard |
| 7 | | 70.7% 67.9 to 73.3 | Reasoning: model default (thinks) | Apr 2025 | Official leaderboard |
| 8 | | 68.1% 65.1 to 70.9 | Reasoning: model default (thinks) | Aug 2026 | Official leaderboard |
| 9 | | 66.6% ±2.9 | Reasoning: model default (thinks) | Aug 2026 | Official leaderboard |
Rankings by part
We read 12 parts from the benchmark’s paper in 3 groups: t1 macro-f1, multilingual evidence and invalid outputs. Each ranks models on one part of the test. Gemini 3.8 Flash leads 6 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
EN
9 models we track have a score here.
- 1
Gemini 3.8 Flash 95.9%
- 2
Gemma 4 31B IT 90.6%
- 3
Qwen3.5-122B-A10B ↑2 87.8%
- 4
DeepSeek V4.1 Flash 87.7%
- 5 GPT-5.6 Luna ↓2 87.2%
- 6 Claude Haiku 4.5 84.6%
- 7
Qwen3-8B 72.3%
- 8
Qwen 3.8 27B 72.3%
- 9
Qwen3.8-Flash-Next 69.7%
FR
9 models we track have a score here.
- 1
Gemini 3.8 Flash 95.1%
- 2
Gemma 4 31B IT 89.2%
- 3 GPT-5.6 Luna 88.5%
- 4
DeepSeek V4.1 Flash 86.9%
- 5 Claude Haiku 4.5 ↑1 86.4%
- 6
Qwen3.5-122B-A10B ↓1 86.2%
- 7
Qwen3-8B 73.1%
- 8
Qwen 3.8 27B 63.4%
- 9
Qwen3.8-Flash-Next 62.1%
EL
9 models we track have a score here.
- 1
Gemini 3.8 Flash 95.1%
- 2
Gemma 4 31B IT 93.1%
- 3
DeepSeek V4.1 Flash ↑1 88.1%
- 4
Qwen3.5-122B-A10B ↑1 85.9%
- 5 GPT-5.6 Luna ↓2 84.7%
- 6 Claude Haiku 4.5 83.7%
- 7
Qwen 3.8 27B ↑1 75.5%
- 8
Qwen3.8-Flash-Next ↑1 74.1%
- 9
Qwen3-8B ↓2 61.9%
Text
9 models we track have a score here.
- 1
Gemini 3.8 Flash 86.9%
- 2
Qwen 3.8 27B ↑6 86.5%
- 3
Qwen3.8-Flash-Next ↑6 84.1%
- 4
Qwen3.5-122B-A10B ↑1 83.5%
- 5
Gemma 4 31B IT ↓3 75.4%
- 6 GPT-5.6 Luna ↓3 73.6%
- 7
DeepSeek V4.1 Flash ↓3 72.4%
- 8 Claude Haiku 4.5 ↓2 71.9%
- 9
Qwen3-8B ↓2 62%
T2a accuracy
9 models we track have a score here.
- 1
Gemini 3.8 Flash 99.4%
- 2
Qwen3.5-122B-A10B ↑3 99%
- 3
Gemma 4 31B IT ↓1 98.8%
- 4
Qwen 3.8 27B ↑4 98.5%
- 5
Qwen3.8-Flash-Next ↑4 98.5%
- 6
DeepSeek V4.1 Flash ↓2 98.3%
- 7 Claude Haiku 4.5 ↓1 97.5%
- 8 GPT-5.6 Luna ↓5 97.3%
- 9
Qwen3-8B ↓2 94.2%
T2b macro-F1
9 models we track have a score here.
- 1
Qwen3.5-122B-A10B ↑4 90.4%
- 2
Gemini 3.8 Flash ↓1 89.2%
- 3 GPT-5.6 Luna 88.3%
- 4
Gemma 4 31B IT ↓2 87.4%
- 5 Claude Haiku 4.5 ↑1 86.2%
- 6
DeepSeek V4.1 Flash ↓2 86.2%
- 7
Qwen3-8B 80.4%
- 8
Qwen3.8-Flash-Next ↑1 73.7%
- 9
Qwen 3.8 27B ↓1 62.5%
T2b exact set
9 models we track have a score here.
- 1
Qwen3.5-122B-A10B ↑4 71.8%
- 2
Qwen3.8-Flash-Next ↑7 67.4%
- 3 GPT-5.6 Luna 65.8%
- 4
Gemini 3.8 Flash ↓3 65.6%
- 5
Gemma 4 31B IT ↓3 65.1%
- 6 Claude Haiku 4.5 60.1%
- 7
DeepSeek V4.1 Flash ↓3 59.7%
- 8
Qwen 3.8 27B 57%
- 9
Qwen3-8B ↓2 45.5%
T2c accuracy
9 models we track have a score here.
- 1
Gemini 3.8 Flash 95.1%
- 2
Qwen3.5-122B-A10B ↑3 91.6%
- 3
Gemma 4 31B IT ↓1 89%
- 4
Qwen3.8-Flash-Next ↑5 75.5%
- 5
Qwen 3.8 27B ↑3 72.3%
- 6
Qwen3-8B ↑1 68.7%
- 7 Claude Haiku 4.5 ↓1 59.9%
- 8 GPT-5.6 Luna ↓5 49.4%
- 9
DeepSeek V4.1 Flash ↓5 35.6%
T1
9 models we track have a score here.
- 1
Qwen3-8B ↑6 0%
- 2 Claude Haiku 4.5 ↑4 0%
- 3
Gemma 4 31B IT ↓1 0%
- 4 GPT-5.6 Luna ↓1 0%
- 5
DeepSeek V4.1 Flash ↓1 0%
- 6
Gemini 3.8 Flash ↓5 1.00%
- 7
Qwen3.5-122B-A10B ↓2 6.00%
- 8
Qwen 3.8 27B 31%
- 9
Qwen3.8-Flash-Next 33%
T2a
9 models we track have a score here.
- 1
Qwen3-8B ↑6 0%
- 2 Claude Haiku 4.5 ↑4 0%
- 3
Qwen3.5-122B-A10B ↑2 0%
- 4
Gemma 4 31B IT ↓2 0%
- 5 GPT-5.6 Luna ↓2 0%
- 6
Gemini 3.8 Flash ↓5 0%
- 7
DeepSeek V4.1 Flash ↓3 0%
- 8
Qwen 3.8 27B 1.00%
- 9
Qwen3.8-Flash-Next 2.00%
T2b
9 models we track have a score here.
- 1
Qwen3-8B ↑6 0%
- 2 Claude Haiku 4.5 ↑4 0%
- 3
Qwen3.5-122B-A10B ↑2 0%
- 4
Gemma 4 31B IT ↓2 0%
- 5 GPT-5.6 Luna ↓2 0%
- 6
Gemini 3.8 Flash ↓5 0%
- 7
DeepSeek V4.1 Flash ↓3 0%
- 8
Qwen3.8-Flash-Next ↑1 23%
- 9
Qwen 3.8 27B ↓1 34%
T2c
9 models we track have a score here.
- 1
Qwen3-8B ↑6 0%
- 2 Claude Haiku 4.5 ↑4 0%
- 3
Gemma 4 31B IT ↓1 0%
- 4 GPT-5.6 Luna ↓1 0%
- 5
Gemini 3.8 Flash ↓4 0%
- 6
Qwen3.5-122B-A10B ↓1 1.00%
- 7
DeepSeek V4.1 Flash ↓3 8.00%
- 8
Qwen3.8-Flash-Next ↑1 23%
- 9
Qwen 3.8 27B ↓1 27%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MultiFinGround paper
Claude Haiku 4.5 on the official leaderboard passed 85% in October 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MultiFinGround 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
8 of 9 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (15 entries).
- Pass Models scored 9 models we track, plus 6 other entries on its leaderboard.
- Pass Newest score The newest is from September 2026, for DeepSeek V4.1 Flash.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Gemini 3.8 Flash (#24).
- Note Headroom Saturated. The best is 95.4%, past 85% since October 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 9 rows for models we track and 6 other entries 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.