AI-Search vs Voyage AI

Voyage AI's capability set is entirely about retrieval quality. The lineup spans general-purpose embedders (voyage-3.5, 3.5 lite), domain-tuned models for finance, legal and code, and company-specific fine-tuning on proprietary data — the fine-tuned legal embedder cited in the Harvey case cut irrelevant documents by 25% and vectorDB costs threefold. The Voyage 4 series and voyage-multimodal-3.5 extend retrieval past plain text so images and text run through one pipeline. Engineering-level features matter here: 32K-token support for embedding long documents, low-dimensional embeddings (3x-8x shorter vectors) that cut storage and search cost, voyage-context-3 pairing chunk-level detail with global document context, a Batch API for large-scale embedding jobs, and rerank-2.5 / rerank-2.5-lite adding instruction following so ranking behavior is steerable in plain language. AI-Search shares none of that surface area. It is a reading tool: 42 sources indexed in Chinese, per-article AI 点评, a weekly AI 总结, daily 日报 and weekly 周报, AI 每日必读 picks, per-source counters with stalled-source flagging, category filters, a ⌘K command palette, time filters (24h/3d/7d/30d), GitHub-star and Hacker News heat rankings, and read/unread plus saved collections. It stores only headlines, short summaries and links — no full article text, no original reporting.

Voyage AI is a paid product with consumption-based pricing, and its own positioning flags the risk: workloads too spiky for consumption pricing can get uneconomical, and small teams still testing whether RAG works are explicitly not the target. The economics are argued through efficiency rather than list price — 3x-8x shorter embeddings reduce vector storage and search bills, and the reranker line is pitched at "2x cheaper inference with superior accuracy" alongside "4x smaller model with faster inference." The Batch API exists to make large embedding workloads cheaper and more efficient. What the data provided does not give you is a dollar-per-token figure or plan tiers — check current rates before budgeting. AI-Search sits at the opposite extreme: free, browser-based, no login. There is no tier to upgrade, no seat count, no metered usage. For a buyer, the comparison is really budget shape: Voyage AI is an ongoing inference line item that scales with how much you embed and rerank, justified by retrieval accuracy on documents where being wrong is expensive. AI-Search has zero procurement friction and zero cost, but also zero integration surface — it's a site you open.