Voyage Finance 2
Voyage Finance 2 is Voyage AI's finance-specialized embedding model with a context window of 0 tokens. It achieves 0.831 average NDCG@10 across 11 financial retrieval datasets, outperforming OpenAI text-embedding-3-large by 7% and Cohere Embed v3 by 12%.
import { embed } from 'ai';
const result = await embed({ model: 'voyage/voyage-finance-2', value: 'Sunny day at the beach',})Frequently Asked Questions
What types of financial documents does Voyage Finance 2 handle?
Corporate event summaries, public company filings, 10-K reports, tabular financial data, personal finance content, and documents requiring numerical reasoning across tables and text.
How does Voyage Finance 2 compare to general-purpose embedding models on finance?
Voyage Finance 2 outperforms OpenAI text-embedding-3-large by 7% and Cohere Embed v3 by 12% across 11 financial retrieval datasets, with an average NDCG@10 of 0.831.
Should I use Voyage Finance 2 or voyage-3-large for financial retrieval?
Voyage AI's voyage-3-large now outperforms domain-specific models on financial benchmarks. For mixed financial and non-financial content, voyage-3-large or voyage-3.5 may be simpler. Use Voyage Finance 2 if your workload is exclusively financial and you have existing indices.
What is the context window for Voyage Finance 2?
0 tokens. This handles lengthy financial filings and multi-page research reports without truncation, which is critical for documents where key disclosures appear deep in the text.
Does Voyage Finance 2 handle tables and numerical data?
Yes. Voyage Finance 2 shows particular strength on benchmarks involving hybrid tabular and textual financial data (TAT-QA, ConvFinQA, FinQA), which test retrieval requiring numerical reasoning.
How do I authenticate Voyage Finance 2 through Vercel AI Gateway?
Add your Voyage AI API key in AI Gateway settings, then send embedding requests through AI Gateway. AI Gateway authenticates requests and records embedding usage.
What financial benchmarks is Voyage Finance 2 evaluated on?
11 financial retrieval datasets including TAT-QA (0.788 NDCG@10), ConvFinQA (0.820), and FinQA (0.795). These cover corporate filings, financial news, tabular data, and numerical reasoning tasks.