Organizations: South China University of Technology · National University of Singapore · iFLYTEK CO., LTD · Central University of Finance and Economics · 6Estates · Hefei University of Technology
Deep Research (DR) agents have demonstrated strong capabilities in complex, research-oriented tasks through autonomous planning, iterative retrieval, multi-step reasoning, and structured reporting. However, adapting DR agents to finance introduces unique challenges: financial analysis demands the joint completion of heterogeneous sub-tasks spanning diverse data types, tools, and analytical workflows. We identify three key requirements for a professional financial DR agent: integration of authoritative, heterogeneous financial data sources; specialized analytical tools and skills; and dedicated sub-agents for domain-specific sub-tasks. Building on these principles, we propose FinNextAssist, an end-to-end deep research framework designed for professional financial analysis. FinNextAssist decomposes the research process into four stages: Task Planner, Evidence Compiler, Reasoning Engine, and Report Assembler, and introduces two novel lightweight sub-agents: TabAgent, for cross-market financial table understanding, and HeteroAgent, for cross-modality heterogeneous financial data interpretation. Extensive experiments on FinDeepResearch, the Finance Agent Benchmark, and FinTMMBench-Web show that FinNextAssist substantially outperforms both strong proprietary and open-source DR agents, with ablation studies confirming the contribution of each component across diverse markets and languages.
Figures & tables
Figure 1. Representative failures of existing Deep Research agents in financial analysis.
Figure 2. Overview of the FinNextAssist framework.
Figure 3. Information Precision of all evaluated methods on the FinDeepResearch benchmark.
Figure 4. Performance analysis across four different capabilities.
Method
Accuracy (%)
Δ vs. FinNextAssist
DeepSeek-v3.2
6.18
− 71.21
Gemini 2.5 Pro
44.53
− 32.86
Grok 4
51.85
− 25.54
OpenAI GPT-5
53.02
− 24.37
Claude-Sonnet-4.5
56.71
− 20.68
FinNextAssist
77.39
–
Table 1. Performance analysis on Finance Agent Benchmark.
Method
Retrieval Corpus
EM
F1
Acc.
LLM Acc.
GPT-4o-mini
Local
4.51
5.89
6.68
6.25
BM25
Local
10.85
20.89
15.89
11.97
BGE-Text
Local
17.11
27.36
23.41
18.50
BGE-Visual
Local
14.51
25.45
21.89
16.04
TMMHybridRAG
Local
19.12
31.41
26.56
21.53
DeepSeek-v3.2
Web
6.84
10.78
7.80
12.67
Table 2. Performance on FinTMMBench-Web. Bold marks the best result overall; underline marks the best Web-retrieval baseline.
Method
S1
S2
S3
S4
S5
S6
LLM (Thinking)
Gemini 2.5 Pro (T)
39.1
15.2
32.0
37.3
44.0
14.5
DeepSeek-v3.2 (T)
37.4
14.6
28.8
33.6
36.8
12.8
Claude-Sonnet-4.5 (T)
42.0
13.0
36.1
47.7
47.3
13.3
Grok 4 (T)
42.8
19.2
30.1
36.9
37.7
13.9
OpenAI GPT-5 (T)
41.7
12.8
29.5
40.5
41.8
12.6
Table 3. Performance analysis across six sections. Values denote normalized accuracy scores. The best and second-best methods are indicated with bold and underline .
Method
US
UK
CN
HK
AU
SG
MY
ID
LLM (Thinking)
Gemini 2.5 Pro (T)
19.9
21.0
17.6
20.8
24.4
24.2
25.1
16.5
DeepSeek-v3.2 (T)
19.7
17.7
17.3
18.4
20.9
21.0
23.8
15.0
Claude-Sonnet-4.5 (T)
22.2
19.9
19.1
21.7
23.0
22.7
24.7
17.0
Grok 4 (T)
23.2
24.0
16.9
18.4
25.8
24.3
25.0
17.4
Table 4. Performance analysis across eight financial markets. Values denote normalized accuracy scores. The best and second-best methods are indicated with bold and underline .
Agent Setting
Overall
Capability Score
Tab Agent
Hetero Agent
Score
Δ
Recog.
Calc.
Abstr.
Interp.
✗
✗
42.7
-5.8
66.5
55.0
41.4
20.6
✗
✓
44.2
-4.3
66.1
66.4
41.6
20.7
✓
✗
46.0
-2.5
75.9
61.1
41.5
20.7
✓
✓
48.5
–
76.1
75.0
42.9
20.9
Table 5. Ablation study of TabAgent and HeteroAgent.
LLM Backbone
Overall
Capability Score
Score
Δ
Recog.
Calc.
Abstr.
Interp.
GPT-4o-mini
46.8
-1.7
75.8
67.5
41.4
20.4
DeepSeek-v3.2
47.1
-1.4
76.2
68.9
41.6
20.3
Claude-Sonnet-4.5
48.5
–
76.1
75.0
42.9
20.9
Table 6. Ablation study of different LLM backbones.
School of Computing, National University of Singapore · Institute of Computing Technology, Chinese Academy of Sciences · China Economics and Management Academy, Central University of Finance and Economics +1