Benchmark Almanac

FinHopper

Claude Sonnet 4.5 67.3% leads since October 2026
Not in model cards

Agents research financial facts through up to six dependent searches. Intermediate answer keys show where an incorrect company, date, or indicator derails the search.

Released
October 2026
Kind
Finance
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Actions, Text
Interaction
Agentic
Environment
APIs and tools
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-5.2 #51 on the capability index
Last model added
December 2025
Scores refreshed
November 20th, 2026
12
Entries in its paper, 9 models we track

Claude Sonnet 4.5 leads with 67.3% among models released by December 2025 Claude Sonnet 4.5 leads with 67.3% among models released by December 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. 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 December 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude Sonnet 4.5 thinking, use tools 67.3%
  2. 2 Gemini 3 Pro thinking, use tools 65%
  3. 3 GPT-5.2 thinking, use tools 61.3%
  4. 4 Seed1.8 thinking, use tools 61.1%
  5. 5 DeepSeek-V3.2 thinking, use tools 58.1%
  6. 6 Kimi K2 Thinking thinking, use tools 57.5%
  7. 7 HunYuan-2.0 thinking, use tools 57.3%
  8. 8 Kimi K2 Sep 2025, use tools 54.6%
  9. 9 HunYuan-2.0-instruct use tools 52.7%
  10. 10 Grok 4.1 thinking, use tools 52.7%
  11. 11 Qwen-max-20250923 thinking, use tools 52.5%
  12. 12 GLM-4.7 thinking, use tools 49%

Source: The FinHopper paper

#EntryScoreSettingDateSource
1 Claude Sonnet 4.5 67.3% thinking, use tools Sep 2025 Official leaderboard
2 Gemini 3 Pro 65% thinking, use tools Nov 2025 Official leaderboard
3 GPT-5.2 61.3% thinking, use tools Dec 2025 Official leaderboard
4 Seed1.8 61.1% thinking, use tools Dec 2025 Official leaderboard
5 DeepSeek-V3.2 58.1% thinking, use tools Dec 2025 Official leaderboard
6 Kimi K2 Thinking 57.5% thinking, use tools Nov 2025 Official leaderboard
7 HunYuan-2.0 57.3% thinking, use tools Official leaderboard
8 Kimi K2 Sep 2025 54.6% use tools Sep 2025 Official leaderboard
9 HunYuan-2.0-instruct 52.7% use tools Official leaderboard
10 Grok 4.1 52.7% thinking, use tools Nov 2025 Official leaderboard
11 Qwen-max-20250923 52.5% thinking, use tools Official leaderboard
12 GLM-4.7 49% thinking, use tools Dec 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Sonnet 4.5 67.3% thinking, use tools Sep 2025 Official leaderboard
2 Gemini 3 Pro 65% thinking, use tools Nov 2025 Official leaderboard
3 GPT-5.2 61.3% thinking, use tools Dec 2025 Official leaderboard
4 Seed1.8 61.1% thinking, use tools Dec 2025 Official leaderboard
5 DeepSeek-V3.2 58.1% thinking, use tools Dec 2025 Official leaderboard
6 Kimi K2 Thinking 57.5% thinking, use tools Nov 2025 Official leaderboard
7 Kimi K2 Sep 2025 54.6% use tools Sep 2025 Official leaderboard
8 Grok 4.1 52.7% thinking, use tools Nov 2025 Official leaderboard
9 GLM-4.7 49% thinking, use tools Dec 2025 Official leaderboard

Rankings by part

We read 7 parts from the benchmark’s paper, each ranking models on one part of the test, such as 1hop, 2hop and 3hop. Claude Sonnet 4.5 leads 4 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

1hop

9 models we track have a score here.

  1. 1 DeepSeek-V3.2 ↑4 88.5%
  2. 2 GPT-5.2 ↑1 86.8%
  3. 3 Claude Sonnet 4.5 ↓2 81.3%
  4. 4 Gemini 3 Pro ↓2 81.2%
  5. 5 Seed1.8 ↓1 79%
  6. 6 GLM-4.7 ↑3 76.1%
  7. 7 Kimi K2 Sep 2025 68%
  8. 8 Kimi K2 Thinking ↓2 67.2%
  9. 9 Grok 4.1 ↓1 66.2%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Sonnet 4.5
no-thinking… no-thinking…
67.3%
Gemini 3 Pro
no-thinking… no-thinking…
65%
GPT-5.2
without too… use tools
61.3%
Seed1.8
thinking, w… use tools
61.1%
DeepSeek-V3.2
thinking, w… no-thinking…
58.1%
Kimi K2 Thinking
thinking, w… thinking, u…
57.5%
Kimi K2
without too… use tools
54.6%
Grok 4.1
thinking, w… use tools thinking, u…
52.7%
GLM-4.7
no-thinking… no-thinking…
49%

Source: The FinHopper paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Sonnet 4.5 thinking, use tools 67.3%
Gemini 3 Pro thinking, use tools 65%
Claude Sonnet 4.5 no-thinking, use tools 63.7%
Gemini 3 Pro no-thinking, use tools 62.7%
GPT-5.2 thinking, use tools 61.3%
Seed1.8 thinking, use tools 61.1%
Seed1.8 use tools 59.2%
GPT-5.2 use tools 58.5%
DeepSeek-V3.2 thinking, use tools 58.1%
Kimi K2 Thinking thinking, use tools 57.5%
DeepSeek-V3.2 no-thinking, use tools 57.1%
Kimi K2 Sep 2025 use tools 54.6%
Grok 4.1 thinking, use tools 52.7%
GLM-4.7 thinking, use tools 49%
GLM-4.7 no-thinking, use tools 48.8%
Grok 4.1 use tools 42.7%
GLM-4.7 thinking, without tools 8.10%
Kimi K2 Sep 2025 without tools 7.00%
GLM-4.7 no-thinking, without tools 6.40%
Grok 4.1 without tools 6.20%
DeepSeek-V3.2 no-thinking, without tools 4.20%
GPT-5.2 thinking, without tools 4.20%
Kimi K2 Thinking thinking, without tools 3.40%
Gemini 3 Pro thinking, without tools 3.20%
Seed1.8 without tools 3.20%
Grok 4.1 thinking, without tools 3.00%
Gemini 3 Pro no-thinking, without tools 2.80%
Claude Sonnet 4.5 thinking, without tools 2.50%
DeepSeek-V3.2 thinking, without tools 2.00%
Seed1.8 thinking, without tools 1.80%
GPT-5.2 without tools 1.40%
Claude Sonnet 4.5 no-thinking, without tools 1.00%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Sonnet 4.5
no-thinking… no-thinking…
67.3%
Gemini 3 Pro
no-thinking… no-thinking…
65%
GPT-5.2
without too… use tools
61.3%
Seed1.8
thinking, w… use tools
61.1%
DeepSeek-V3.2
thinking, w… no-thinking…
58.1%
Kimi K2 Thinking
thinking, w… thinking, u…
57.5%
Kimi K2
without too… use tools
54.6%
Grok 4.1
thinking, w… use tools thinking, u…
52.7%
GLM-4.7
no-thinking… no-thinking…
49%

Source: The FinHopper paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Sonnet 4.5 thinking, use tools 67.3%
Gemini 3 Pro thinking, use tools 65%
Claude Sonnet 4.5 no-thinking, use tools 63.7%
Gemini 3 Pro no-thinking, use tools 62.7%
GPT-5.2 thinking, use tools 61.3%
Seed1.8 thinking, use tools 61.1%
Seed1.8 use tools 59.2%
GPT-5.2 use tools 58.5%
DeepSeek-V3.2 thinking, use tools 58.1%
Kimi K2 Thinking thinking, use tools 57.5%
DeepSeek-V3.2 no-thinking, use tools 57.1%
Kimi K2 Sep 2025 use tools 54.6%
Grok 4.1 thinking, use tools 52.7%
GLM-4.7 thinking, use tools 49%
GLM-4.7 no-thinking, use tools 48.8%
Grok 4.1 use tools 42.7%
GLM-4.7 thinking, without tools 8.10%
Kimi K2 Sep 2025 without tools 7.00%
GLM-4.7 no-thinking, without tools 6.40%
Grok 4.1 without tools 6.20%
DeepSeek-V3.2 no-thinking, without tools 4.20%
GPT-5.2 thinking, without tools 4.20%
Kimi K2 Thinking thinking, without tools 3.40%
Gemini 3 Pro thinking, without tools 3.20%
Seed1.8 without tools 3.20%
Grok 4.1 thinking, without tools 3.00%
Gemini 3 Pro no-thinking, without tools 2.80%
Claude Sonnet 4.5 thinking, without tools 2.50%
DeepSeek-V3.2 thinking, without tools 2.00%
Seed1.8 thinking, without tools 1.80%
GPT-5.2 without tools 1.40%
Claude Sonnet 4.5 no-thinking, without tools 1.00%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Benchmark paper Not tested since Oct 2026

Source: The FinHopper paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Sonnet 4.5 Kimi K2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Sonnet 4.5 Kimi K2

Source: Official FinHopper 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

All 637 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

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 11 months ago, in December 2025, for GLM-4.7. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (12 entries).
  • Pass Models scored 9 models we track, plus 3 other entries on its leaderboard.
  • Warning Newest score The newest came 11 months ago, in December 2025, for GLM-4.7.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.2 (#51).
  • Pass Headroom The best is 67.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 32 rows for models we track and 3 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.

How we check