Best AI for Filings Q&A (November 2026)
Data refreshed 1 Nov 2026
Claude Opus 5 leads CorpFin v2 with 73.2 (accuracy %). GLM-5.2 leads IPO Finance Agent with 79.8 (mean rubric score %). o1 leads FinanceQA with 0.487 (accuracy (total, 0-1)). GPT-4-Turbo leads FinanceBench with 85 (% correct (150-question sample)).
- Benchmarks in this use case
- 4
- With dated scores
- 4
- Models with scores
- 23
- Latest source date
- 12 Aug 2026
Filings Q&A picks
The leader on each benchmark, never combinedBest on CorpFin v2
Claude Opus 5Anthropic
73.2 accuracy %
Official board, 12 Aug 2026
Best on IPO Finance Agent
GLM-5.2Zhipu AI
79.8 mean rubric score %
Reported by the authors, 30 Jun 2026
Best on FinanceQA
o1OpenAI
0.487 accuracy (total, 0-1)
Reported by the authors, 30 Jan 2025
Best on FinanceBench
GPT-4-TurboOpenAI
85 % correct (150-question sample)
Reported by the authors, 20 Nov 2023
Filings Q&A leaderboard
CorpFin v2 v2 (archived; last updated 2026-08-12)
accuracy %, higher is better. Top 5 of 10 models.
Updated 12 Aug 2026
- 1Claude Opus 573.2
- 2Claude Fable 571.8
- 3Kimi K371.6
- 4Muse Spark 1.171.3
- 5Muse Spark 1.270.9
| 1 | Claude Opus 5 | Anthropic | 73.2 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
|---|---|---|---|---|---|
| 2 | Claude Fable 5 | Anthropic | 71.8 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 3 | Kimi K3 | Moonshot AI | 71.6 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 4 | Muse Spark 1.1 | Not stated | 71.3 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 5 | Muse Spark 1.2 | Not stated | 70.9 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 6 | Inkling Small | Not stated | 69.6 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 7 | Inkling | Not stated | 68.6 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 8 | Grok 4.3 | xAI | 68.5 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 9 | GPT 5.5 | OpenAI | 68.4 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
| 10 | Kimi K2.5 | Moonshot AI | 68.3 | Default | Vals.ai leaderboard (archived) Official board, 12 Aug 2026 |
IPO Finance Agent arXiv v3 (2026-06-30)
mean rubric score %, higher is better. Top 5 of 6 models.
Updated 30 Jun 2026
- 1GLM-5.279.8
- 2GLM-5.179.7
- 3Qwen 3.7 Max78.9
- 4MiMo-2.5 Pro77.2
- 5Kimi K2.668
| 1 | GLM-5.2 | Zhipu AI | 79.8 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
|---|---|---|---|---|---|
| 2 | GLM-5.1 | Zhipu AI | 79.7 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
| 3 | Qwen 3.7 Max | Qwen | 78.9 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
| 4 | MiMo-2.5 Pro | Xiaomi | 77.2 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
| 5 | Kimi K2.6 | Moonshot AI | 68 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
| 6 | Nemotron 3 Ultra | NVIDIA | 45.5 | public SpaceX set (n=70), agent harness | Paper Table 1 Reported by the authors, 30 Jun 2026 |
FinanceQA v1 (2025-01-30)
accuracy (total, 0-1), higher is better. 4 models.
Updated 30 Jan 2025
- 1o10.487
- 2Claude-3-5-sonnet-202410220.399
- 3GPT-4o0.392
- 4Llama-3.3-70B-Instruct0.311
| 1 | o1 | OpenAI | 0.487 | Default | Paper Table 1 Reported by the authors, 30 Jan 2025 |
|---|---|---|---|---|---|
| 2 | Claude-3-5-sonnet-20241022 | Anthropic | 0.399 | Default | Paper Table 1 Reported by the authors, 30 Jan 2025 |
| 3 | GPT-4o | OpenAI | 0.392 | Default | Paper Table 1 Reported by the authors, 30 Jan 2025 |
| 4 | Llama-3.3-70B-Instruct | Meta | 0.311 | Default | Paper Table 1 Reported by the authors, 30 Jan 2025 |
FinanceBench arXiv v1 (2023-11-20); HF dataset last modified 2024-11-17
% correct (150-question sample), higher is better. 3 models.
Updated 20 Nov 2023
- 1GPT-4-Turbo85
- 2Claude276
- 3Llama241
| 1 | GPT-4-Turbo | OpenAI | 85 | Oracle (evidence pages given) | Paper Table 2 Reported by the authors, 20 Nov 2023 |
|---|---|---|---|---|---|
| 2 | Claude2 | Anthropic | 76 | Long context (filing in context) | Paper Table 2 Reported by the authors, 20 Nov 2023 |
| 3 | Llama2 | Meta | 41 | Single vector store (one per filing) | Paper Table 2 Reported by the authors, 20 Nov 2023 |