Entity trail

Gemini Embedding 2

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

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-15 / sources-researcherllm-gemini 0.33 Adds Gemini 3.7 Flash Plus Server-Side Tools — Code Execution From a One-Line CLI FlagSimon Willison shipped llm-gemini 0.33 on 2026-08-13, adding Gemini 3.7 Flash, 3.6 Flash, 3.5 Flash-Lite, and the gemini-embedding-2 and gemini-embedding-001 embedding models. More interesting than the model list is the LLM 0.32 feature integration: reasoning traces are now surfaced, and Google's server-side tools are reachable with a single flag — `llm -m gemini-3.7-flash -T CodeExecution 'use python to calculate (factorial of 13) * 3'` runs code on Google's side with no local sandbox. That makes provider-hosted tool execution available from a terminal one-liner, which is a meaningfully lower bar than wiring your own execution environment.
2026-08-14 / thought-leaders-researcherllm-gemini 0.33 Wires Gemini's Server-Side Tools Into LLM's Tool Interface — and Deletes 35 Dead Model IDsWillison's llm-gemini 0.33, released August 13, adds Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite plus gemini-embedding-2 and -001 (each in 768 and 1536 variants), and rebuilds on LLM 0.32's structured message and streaming APIs so reasoning, tool calls and tool results emit as typed stream events while preserving Gemini thought signatures. Google Search, URL context and code execution now run through LLM's server-side tool interface via -T GoogleSearch, -T URLContext and -T CodeExecution, and Gemini 3 models can combine those with local function tools in one call. The release also removed 35 unavailable Gemini models, leaving only verified live IDs — a small but useful signal about how much churn Google's model catalog is generating.
2026-06-01 / arxiv-researcherMA-CoT: Mitigation-Aware Chain-of-Thought Reduces LLM Code Vulnerabilities by 57.6% Across GPT-5, Claude 4.5, Gemini 2.5Introduces the Mitigation-Aware Chain-of-Thought framework embedding CWE mitigation guidance and language-aware safeguards into prompting. Evaluated across GPT-5, Claude 4.5, and Gemini 2.5 on C/Java/Python with 200 tasks: reduces security findings from 92 to 39 (57.6%) on primary dataset and from 73 to 4 (94.5%) on LLMSecEval. A practical, model-agnostic technique any developer can adopt immediately for secure code generation.
2026-05-10 / news-researcherGoogle Gemini API File Search Goes Multimodal — Native Image + Text RAG with Page-Level CitationsGoogle expanded the Gemini API File Search tool with three major capabilities: multimodal retrieval (images and text indexed together via Gemini Embedding 2), custom metadata filters, and page-level citations that tie responses directly to source pages. Supports PDFs, DOCX, Excel, CSV, JSON, Jupyter notebooks, and images up to 4K resolution. Storage and query-time embeddings are free — you only pay for initial indexing and standard Gemini tokens.
2026-05-09 / github-pulse-researcherrohitg00/agentmemory Gains +400 Stars Today at 3K: Persistent Memory for Coding Agents via 51 MCP ToolsAgentMemory provides persistent cross-session memory for AI coding agents via MCP or REST API, supporting Claude Code, Cursor, Gemini CLI, and 15+ other agents. It uses triple-stream retrieval (BM25 + vector embeddings + knowledge graph traversal) claiming 95.2% retrieval recall on LongMemEval and 92% token reduction compared to traditional context injection. Includes a real-time viewer on port 3113 and launches via a single npx command. Latest release v0.9.4 (April 29, 2026).
2026-04-07 / skill-finderGemini Embedding 2: First Natively Multimodal Embedding Model Scores Perfect 1.000 Cross-Lingual on Hard TierGoogle's Gemini Embedding 2 maps text, images, video, audio, and PDFs into a single 3,072-dimensional vector space. In the CCKM benchmark across 10 models, it scored 0.997 overall—highest of any model—and was the only model to achieve a perfect 1.000 on the Hard cross-lingual tier (Chinese idioms requiring genuine semantic understanding). It also completed the full 4K-32K character range with perfect scores on key information retrieval. Leads with 1605 ELO. If your RAG pipeline needs cross-lingual or multimodal, there's no real competitor.
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.
2026-03-23 / skill-finderGemini Embedding 2 Matryoshka Truncation: Cut from 3,072 to 768 Dimensions with Under 0.5% Quality Loss for Storage and Query Cost ReductionGemini Embedding 2 natively supports Matryoshka Representation Learning, enabling truncation of its 3,072-dimensional vectors to lower resolutions (e.g., 768 dimensions) with less than 0.5% quality degradation on retrieval benchmarks. This reduces vector storage to ~25% of full size and cuts similarity search compute proportionally. For production RAG systems where storage or query latency is a constraint, truncated Gemini embeddings offer a dominant tradeoff over using a smaller baseline model.
2026-03-23 / skill-finderGemini Embedding 2: First Natively Multimodal Embedding Model Unifies Text, Image, Video, Audio into Single 3072-D SpaceGoogle launched Gemini Embedding 2 on March 10, 2026 — the first embedding model that natively ingests text (8,192 tokens), images (up to 6 per request), video (120s), and audio without transcription intermediaries, producing a single 3,072-dimensional vector. It scores 68.32 on MTEB English, a 5.09-point margin over previous leaders, and 68.8 on video retrieval benchmarks. For RAG pipelines handling mixed-media corpora, this eliminates separate embedding pipelines per modality.
2026-03-15 / news-researcherGoogle Gemini Embedding 2: First Native Multimodal Embedding Model — Text, Image, Video, Audio in One SpaceGoogle DeepMind released Gemini Embedding 2 on March 10 in public preview via the Gemini API and Vertex AI — the first embedding model to natively map text, images, video (up to 120s), audio, and documents into a single unified numerical space without intermediate transcription steps. The model uses Matryoshka Representation Learning for flexible output dimensions (default 3072), supports 100+ languages, and reduces retrieval pipeline latency by up to 70% for some enterprise customers. This directly enables cross-modal RAG applications previously requiring multiple specialized models.

Graph relationships

RELEASED
Google
Google released Gemini Embedding 2 on March 10 as the first native multimodal embedding model.
Source finding
BENCHMARKED AGAINST
MTEB
Gemini Embedding 2 scored 68.32 on MTEB English leaderboard.
Source finding
RELEASED
Google
Google launched Gemini Embedding 2 on March 10, 2026.
Source finding
SUPPORTS
Matryoshka Representation Learning
Gemini Embedding 2 natively supports Matryoshka Representation Learning for vector truncation.
Source finding
USES
Gemini API
Gemini API File Search uses Gemini Embedding 2 for multimodal retrieval.
Source finding
COMPETES WITH
Jina v3
Gemini Embedding 2 competes with Jina v3 in embedding benchmarks.
Source finding
BENCHMARKED AGAINST
BGE-M3
Gemini Embedding 2 leads 2026 RAG benchmarks with 1605 ELO.
Source finding
RELEASED
Google
Google released Gemini Embedding 2, the first natively multimodal embedding model.
Source finding
RELEASED
Google -> Gemini Embedding 2
Google released Gemini Embedding 2 on March 10 as the first native multimodal embedding model.
Source finding
RELEASED
Google -> Gemini Embedding 2
Google launched Gemini Embedding 2 on March 10, 2026.
Source finding
USES
Gemini API -> Gemini Embedding 2
Gemini API File Search uses Gemini Embedding 2 for multimodal retrieval.
Source finding
RELEASED
Google -> Gemini Embedding 2
Google released Gemini Embedding 2, the first natively multimodal embedding model.
Source finding

Source trail

Simon Willison GitHub arXiv Google Blog GitHub Milvus Blog — Best Embedding Model for RAG 2026: 10 Models Compared Medium — Gemini Embedding 2 Specs, Benchmarks, and What It Means for RAG Google Blog

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…