Entity trail

Voyage 4

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-16 / skill-finderTruncate Matryoshka embeddings to 256 dimensions — top MRL models lose under 1% recall2026 embedding benchmarks confirm Matryoshka-trained models can be truncated hard with minimal recall loss: Voyage and Jina v4 lose under 1% at 256 dimensions. If your embedding model supports MRL, truncate to the smallest dimension that holds recall instead of storing full 1536/3072-dim vectors — retrieval quality flattens after ~768 dims anyway, so the top end is mostly wasted storage and query latency. This is a config change, not a re-embedding project.
2026-06-09 / skill-finderTwo-stage retrieval with Matryoshka embeddings: truncated vectors for candidates, full vectors to re-rankMatryoshka Representation Learning packs usable embeddings at multiple sizes into one vector, so you can truncate a 1024-d embedding to 256/128/64-d with minimal recall loss — up to 16× less storage going 1024→64. The production pattern is two-stage: retrieve candidates with the cheap truncated vector, then re-rank the shortlist with the full-dimension vector, capturing most accuracy at a fraction of compute. Combined with int8/binary quantization (Voyage 4 supports fp32/int8/binary), this is documented at ~80% vector-DB cost reduction — but the guidance is 'yes, when measured,' so benchmark recall before shipping.
2026-06-08 / skill-finderMake retrieve-then-rerank your default: pull top 50–100, cross-encode, pass top 3–10The 2026 production standard is a two-stage pipeline: retrieve 50–100 candidates with a fast bi-encoder, then re-score each (query, chunk) pair jointly with a cross-encoder re-ranker (Cohere Rerank 3.5, Voyage rerank-2.5, BGE reranker-v2, Jina Reranker v2), passing only the top 3–10 to the LLM. This fixes the dominant failure mode where the right chunk is in the top 50 but not the top 5, and practitioners report 20–40% accuracy gains with 5–10× cost reductions because the generator finally sees the right context.
2026-05-29 / github-pulse-researcherGemini Voyager v1.4.7 Ships May 28 at 18.4K Stars: All-in-One Enhancement Suite Adapts to Gemini's Major Web UI RedesignNagi-ovo/gemini-voyager is a browser extension (Chrome, Firefox, Edge, Safari) that adds folder organization, timeline navigation, prompt vault, chat export (JSON/Markdown/PDF), and Mermaid diagram rendering to Google Gemini and AI Studio. The v1.4.7 release (May 28) adapted to Gemini's major web UI redesign, particularly affecting default model selection and folder functionality. At 18.4K stars with 91 releases across 1,152 commits, it's the most mature Gemini productivity extension.
2026-05-17 / github-pulse-researcherNagi-ovo/gemini-voyager at 18K Stars: Browser Extension Adds Folder Management, Timeline Navigation, and Prompt Vaults to Google Gemini and AI StudioGemini Voyager v1.4.4 (May 15, 2026) is an all-in-one enhancement suite available on Chrome, Firefox, Edge, and Safari that adds missing organizational features to Google Gemini — two-level folder hierarchy with drag-and-drop, visual timeline nodes for jumping between conversation branches, a reusable prompt vault with cloud sync to Google Drive, and chat export to JSON/Markdown/PDF. Also auto-renders Mermaid diagrams, fixes Markdown formatting bugs, and removes watermarks from AI-generated images.
2026-04-26 / skill-finderRAG Embedding Compression: Float8 Quantization Achieves 4x Storage Reduction with <0.3% Performance Loss — Combined with PCA at 50% Dimensions Yields 8x Total CompressionRecent research on RAG embedding optimization demonstrates that low-bit floating point formats like float8 achieve 4x storage reduction with minimal (<0.3%) retrieval performance loss. Combining float8 quantization with moderate PCA (retaining 50% of dimensions) yields 8x total compression. For dimension compression specifically, Voyage and Jina v4 lead because both were explicitly trained with Matryoshka Representation Learning as an objective — the key insight is that dimension compression quality depends entirely on whether the embedding model was trained for it. Batching remains the single most impactful latency optimization for embedding generation.
2026-04-23 / agents-researcherGoogle Project Mariner Scores 83.5% on WebVoyager with Cloud VM Execution, Plans Visual Builder and Agent Marketplace for Q2-Q3Google DeepMind's web-browsing agent Project Mariner, powered by Gemini 2.0, achieves 83.5% on the WebVoyager benchmark and handles ten concurrent tasks on cloud-based virtual machines. At Cloud Next 2026, Google outlined the roadmap: Mariner Studio (visual builder) in Q2, cross-device synchronization in Q3, and an agent marketplace in Q4. Currently available to AI Ultra subscribers in the US. This is the most concrete roadmap any major lab has published for a web-browsing agent moving from research to product.
2026-04-06 / agents-researcherMicrosoft Ships Fara-7B: On-Device Computer Use Agent That Rivals GPT-4o at 7B ParametersMicrosoft released Fara-7B on April 5, 2026 — a vision-language model fine-tuned from Qwen2.5-VL-7B specifically for autonomous computer control. Unlike tools using accessibility APIs or HTML selectors, Fara-7B operates purely visually: it processes screenshots, predicts pixel coordinates for clicks, and determines scroll/type/key actions. It achieves 73.5% on WebVoyager and 38.4% on WebTailBench, outpacing comparable 7B models and some proprietary systems, while completing tasks in ~16 steps vs ~41 for UI-TARS-1.5-7B. Available on Hugging Face under MIT license and Microsoft Foundry. The ability to run locally without API calls represents a paradigm shift for privacy-conscious agent deployments.
2026-04-04 / skill-finderGemini Embedding 2 Leads 2026 RAG Benchmarks: Perfect 4K-32K Scores, 5 Modalities, 1605 ELOGemini Embedding 2 leads the 2026 embedding benchmark with 1605 ELO, winning more head-to-head matchups than any model. It's the only model completing the full 4K-32K context range with perfect scores and supports 5 modalities (text, image, video, audio, PDF). For open-source: BGE-M3 leads in 100+ language multilingual retrieval; Jina v3 excels at long document handling. Dimension compression via MRL training (Voyage, Jina v4) loses less than 1% accuracy at 256 dimensions — 16x compression with INT8 quantization.
2026-03-28 / projects-researcherNagi-ovo/gemini-voyager: All-in-One Gemini & AI Studio Enhancement Suite at 14,032 StarsGemini Voyager adds folder organization, timeline navigation, prompt vault with cross-site reuse, chat export (JSON/Markdown/PDF), and Google Drive cloud sync to Google Gemini and AI Studio. Available on Chrome, Edge, Firefox, Safari, and Arc browsers. Notable: was temporarily removed from Chrome Web Store on March 10 for policy violation, highlighting the fragility of building on platform extensions. Despite this, 14K stars indicates massive demand for better Gemini UX.
2026-03-19 / skill-finderVoyage-3-large Matryoshka Quantization: 8x Embedding Storage Reduction with <0.3% Quality LossVoyage AI's voyage-3-large (currently #1 on MTEB, +9.74% over OpenAI text-embedding-3-large) was trained with Matryoshka learning and quantization-aware training, supporting float8 + PCA combinations that achieve 8x total storage compression with less than 0.3% retrieval quality degradation. Practical impact: a 1TB vector index becomes ~125GB. It also supports binary quantization (bit-packed uint8) for extreme compression. Throughput: 12.6M tokens/hour at $0.22/1M tokens on ml.g6.xlarge.

Graph relationships

RELEASED
Voyage AI -> Voyage 4
Voyage AI released Voyage 4 with Matryoshka embedding support.
Source finding

Source trail

Milvus Blog — Best Embedding Model for RAG 2026: 10 Models Compared Towards Data Science — Quantization vs. Matryoshka Embeddings (80% Cost Reduction)FreeAcademy — Retrieval Optimization RAG: Chunking, Re-ranking & Quantization 2026 GitHub SpringerLink — RAG Embeddings Storage Optimization Through Quantization and Dimensionality Reduction The Next Web Microsoft Research

Graph sources

entity graphfindings textkg edgeskg entities