Gemini

Google's family of foundation models. In this wiki Gemini appears as the provider of Embedding 2, the File Search API, Gemini Flash Live, and now Gemini Robotics-ER 1.6 for embodied reasoning in robot systems.

Key facts

• Gemini Embedding 2 is Google's first natively multimodal embedding model in the RAG cluster, handling text, images, video, audio, and documents in one vector space ^[src-006].
• Gemini 2.5 Flash is the default chat model used against File Search in the earlier RAG workflow ^[src-006].
• Gemini Robotics-ER 1.6 extends the Gemini family into robotics, specializing in spatial reasoning, task planning, success detection, instrument reading, and physical safety constraints ^[src-039].
• Gemini Robotics-ER 1.6 is available through the Gemini API and Google AI Studio ^[src-039].
• In Dwarkesh Patel's Reiner Pope interview, Gemini is used as a pricing and infrastructure example: long-context price tiers and public traffic estimates are treated as clues about Memory Wall For Long Context, Prefill Vs Decode, and model-serving scale ^[src-042].
• Google Cloud Next '26 introduces Gemini 3.1 Pro for complex workflow orchestration, Gemini 3.1 Flash Image / Nano Banana 2 for UI and visual assets, and broader Gemini Enterprise integration across agents, data, Workspace, and customer experience ^[src-044].
• In ^[src-061], Gemini is treated as the main consumer-chatbot challenger to ChatGPT: Gemini 3 had major release momentum and Google has scale, TPUs, and data-center integration, but differentiation and habit still matter against OpenAI's incumbent ChatGPT position.
• In ^[src-062], Pichai frames Gemini as the product/model layer sitting on top of long-term Google bets: TPUs, Brain/DeepMind integration, search, Android/XR, and Google's full-stack infrastructure.
• The same source links Gemini-style multimodal intelligence to Project Astra, Android Xr, AI Mode in Google Search, and Google Beam-style presence products ^[src-062].
• ^[src-105] adds the internal origin story: Gemini is framed as a consolidation of fragmented Google, DeepMind, Pathways, PaLM, and compute efforts into one stronger model effort.
• The same source says Gemini was built around multimodality, tool use, and agentic behavior from early on, with coding and agentic experiences becoming central to the current model cycle ^[src-105].
• ^[src-105] also frames Gemini as a shared intelligence layer that can power many product surfaces while user-facing products remain shaped by human needs, such as search, glasses, audio, and focused task interfaces.
• ^[src-116] connects Gemini Enterprise Agent Platform to Agentic Resource Discovery through Agent Registry, including discoverable resources, trust manifests, egress policies, and Agent Identity.
• ^[src-143] adds Google's API direction: Interactions API is now the primary interface for Gemini models and agents, designed for stateful and long-running workflows.
• ^[src-149] adds another Workspace adoption signal: Gemini in Sheets can diagnose formula errors and propose corrected formulas inside the spreadsheet.

Related concepts

• Embodied Reasoning
• Agentic Vision
• Robotic Success Detection
• Robotic Instrument Reading
• LLM Inference Economics
• Memory Wall For Long Context
• Prefill Vs Decode
• Agentic Enterprise
• Workspace Intelligence
• Enterprise Knowledge Graph
• Model Lab Differentiation
• GPU Supply As AI Strategy
• AI Search Context Layer
• Agentic Operating Systems
• AI Productivity Multiplier
• Agentic AI
• Long Running Agents
• Agent Tool Latency Bottleneck
• Agentic Resource Discovery
• Enterprise Agent Governance
• Agent First Apis
• AI Spreadsheet Assistants

Source references

• ^[src-006] Nate Herk cluster — Nate Herk — RAG and data ingestion cluster (5 videos)

– Videos referenced: hem5D1uvy-w, irg-2IfAjpo

• ^[src-039] Laura Graesser and Peng Xu — "Gemini Robotics-ER 1.6: Powering real-world robotics tasks through enhanced embodied reasoning" (2026-04-14)
• ^[src-042] Dwarkesh Patel — "How GPT, Claude, and Gemini are actually trained and served – Reiner Pope" (2026-04-29)
• ^[src-044] Thomas Kurian — "Welcome to Google Cloud Next '26" (2026-04-22)
• ^[src-061] Lex Fridman – "State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490" (2026-01-31)
• ^[src-062] Lex Fridman – "Sundar Pichai: CEO of Google and Alphabet | Lex Fridman Podcast #471" (2025-06-05)
• ^[src-105] Google for Developers – "Gemini co-leads on project origins and what's next" (2026-05-29)
• ^[src-116] Google Developers Blog – "Announcing the Agentic Resource Discovery specification" (2026-06-17)
• ^[src-143] Ali Cevik / Google – "Interactions API: our primary interface for Gemini models and agents" (2026-06-22)
• ^[src-149] Google Workspace Updates – "Troubleshoot formula errors quickly with Gemini in Google Sheets" (2026-06-22)

2026-06-27 update

• Gemini 3.5 Flash computer use adds a direct UI-action capability to the Gemini watch set ^[src-154].