gemini — LLM AI Model
Gemini answers a prompt with text and reasons well across several steps, many languages, and a large body of factual knowledge. This node is text-only — images and video go to the Nano Banana, VEO, Video Analysis and Omni nodes. Google AI key required.
What it does
Gemini calls Google's hosted models with the text on the node's input port and the instruction you set, and returns generated text. There is no local model and no GPU to provision - the flow talks to Google's API with your own key.
Its tiers span cheap high-throughput variants up to a slower, higher-quality one, and it holds up well on multilingual work and on reasoning that runs over several steps. This is the text node: images and video have their own Gemini nodes, and their ports are where those files go.
When to use it
- Multilingual summarising, translating or rewriting inside a flow
- Turning an earlier node's output into readable text for a person
- High-volume, simple text steps where the cheapest tier is enough
- Reasoning that runs over several steps before producing an answer
Details
- CategoryLLM
- Year2025
- LicenseGoogle Cloud Terms
Compliance & Provenance
- ProviderGoogle · GPAI
- EU AI Act RiskLimited Risk
- Art. 50 TransparencyRequired — AI-generated outputs are marked
Inputs & Outputs
- Text InputInputstring
Input text to process
- Generated TextOutputstring
LLM generated output
Parameters
- modelstringdefault Gemini 3.8 Flashoptions: Gemini 3.8 Flash, Gemini 4 Argon, Gemini 3.7 Flash, +8 more
Model variant
Which Gemini model to call. 'Gemini 3.8 Flash' (default) — newest GA Flash tier, the most capable of them and built for long-horizon work. '3.7 Flash' / '3.6 Flash' / '3.5 Flash' — older GA Flash tiers, still supported. 'Gemini 4 Argon' — Google's flagship reasoning model (GA 2026-10-13), the strongest and by far the costliest here; it always thinks hard before answering (thinking level high), so it is also the slowest. Google serves it on paid-tier API keys only. 'Gemini 3.1 Pro' — high quality, slower and costlier than Flash, still a preview id. 'Gemini 3 Flash' — preview. '3.5 Flash-Lite' / '3.1 Flash-Lite' — cheapest, for high-throughput simple tasks; prefer 3.5 Flash-Lite, since 3.1 Flash-Lite shuts down 2027-05-07. The 2.5 entries are the oldest generation. Output (thinking included) capped by this node's max_tokens (default 16384, max 32768). Pick by quality/cost tradeoff.
- promptstringdefault
System prompt / instruction
Optional system prompt that frames the task (role, style, constraints). The user message comes from the text input pin. Leave blank to use the text input alone.
- max_tokensintegerdefault 16384range 1–32768advanced
Max tokens (1-32768)
Maximum tokens for the response, thinking included: Gemini 3.6 Flash and later think before answering and spend part of this budget doing so, and Gemini 4 Argon thinks the most (often 4000-5000 tokens on a hard problem). Default 16384, maximum 32768. If the answer reaches the limit before it is finished, the node fails with LLM_OUTPUT_TRUNCATED rather than returning a cut-off answer. Lower it only to cap cost, and ask for a shorter answer in the prompt if you need one.
- temperaturefloatdefault 0.7range 0–2advanced
Temperature (0-2, not used by 3.6 Flash and later)
Sampling temperature. 0 = deterministic (best for factual / coding tasks). 0.7 (default) = balanced. >1 = creative. Only the older tiers use it (3.5 Flash, 3.5 Flash-Lite, 3.1 Pro, 3.1 Flash-Lite, 3 Flash, 2.5). Not sent for Gemini 3.6 Flash and later, including the default 3.8 Flash: Google fixes sampling on those models, so changing it has no effect.
Limitations
- Text in, text out. This node has one text input port - for images or video use the Nano Banana, VEO, Video Analysis or Omni nodes.
- Up to 32768 tokens per response, thinking included, so a very long document comes back summarised rather than in full.
- Runs on your own Google AI API key. Without one configured in the workspace the node cannot run.