Loading the catalogue…
Loading the catalogue…
Google's fast, cost-focused Gemini 3.5 model with a million-token context window, mandatory configurable reasoning, and multimodal text, image, audio, and video input.
Overview. Gemini 3.5 Flash-Lite is Google's fast, cost-focused model in the Gemini 3.5 generation, released July 2026 as a closed-weight, proprietary offering. It pairs a 1,048,576-token context window with mandatory, configurable reasoning across four effort levels, and it accepts text, image, audio, and video input while producing text output. Developers reach for it when they need the Gemini Flash family's largest context window at the fastest, most cost-efficient serving tier Google offers.
Key strengths. The model's defining trait is mandatory chain-of-thought reasoning that runs by default at every request, configurable from minimal to high effort, which gives it real multi-step reasoning ability beyond a simple single-turn completion model. It's positioned as the fastest tier in the Gemini 3.x family, built for high-throughput and agentic workloads. Multimodal input spanning text, image, audio, and video broadens what a single Inference call can handle without a separate transcription or vision step.
When to use it. Use Gemini 3.5 Flash-Lite for high-throughput, latency-sensitive agentic tasks of moderate complexity — think tool-heavy pipelines, high-volume classification, or multi-step workflows where speed matters more than maximum reasoning depth. Its default-minimal reasoning effort keeps it responsive out of the box, while the option to raise effort to medium or high gives it headroom for harder steps within the same workflow. The million-token context window makes it a reasonable fit for long-document or long-conversation agentic use cases that would otherwise require chunking.
Limitations. Output is text-only — there's no audio or image generation, so multimodal responses have to be handled downstream. As Google's 'Lite' tier, it sits beneath mainline Flash and Pro variants and defaults to minimal reasoning effort, so the hardest reasoning workloads are better matched to a higher tier. Its training data reaches through approximately March 2026, so very recent events or releases fall outside its knowledge.
Sovereignty note. Gemini 3.5 Flash-Lite is built by Google, a US-headquartered creator, and carries CLOUD Act exposure — meaning US legal authorities can potentially compel access to data processed through it, regardless of where the endpoint is hosted. Teams with strict EU data-residency requirements should evaluate this exposure against their compliance needs rather than treating this model as Sovereign.
| Inference Provider | Sovereignty | Runs in | In / out per 1M | Status | Pin |
|---|---|---|---|---|---|
| L0 | €0.26 / 2.20+3 rates | Online |
18 use cases classified by the curator. The note says why.
Large context window plus configurable reasoning suits grounded answer generation over long retrieved passages at low cost.
Fast, cost-focused tier is well suited to high-throughput classification workloads where latency and unit cost dominate.
Fast, low-cost design with multimodal input (text/image/audio/video) is well matched to high-volume customer support deployments.
Vision capability combined with a 1M-token context window supports OCR-style extraction across long, image-heavy documents.
Million-token context window lets it ingest full document sets for Q&A without chunking, and extended-thinking mode supports multi-hop reasoning over the retrieved content.
Native vision capability handles image-based Q&A directly.
Vision input supports chart interpretation, though no dedicated chart-benchmark evidence is available for this variant.
Fast, low-cost generation is well suited to short-form drafting tasks like email, though not a differentiated strength.
Configurable reasoning and long context support structured entity extraction, though the lite tier trades some accuracy for cost.
Not a dedicated embedding/reranking model, but the huge context window and speed tier make it usable for lightweight search-adjacent tasks.
Extended-thinking mode helps with structured output tasks, but as the cost-tier variant it likely trails larger Gemini 3.5 models on complex schemas.
Native audio input is supported, but this is a general-purpose multimodal model rather than a specialised transcription system.
No tool-calling capability is documented for this variant, and the cost-focused lite tier is not positioned for complex agentic coding workflows.
Sourced numbers, not our own evaluations. Percentiles are against the whole Stav catalogue.
Computed from aa-intelligence-index, gpqa-diamond, aa-engineering-index, aa-coding-index, livecodebench, terminal-bench-v4, aa-lcr.
curl https://api.stav.ai/v1/chat/completions \
-H "Authorization: Bearer $STAV_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemini-3.5-flash-lite",
"messages": [{"role": "user", "content": "Hello!"}]
}'→ X-Stav-Request-Id · X-Stav-Model · X-Stav-Provider → X-Stav-Sovereignty-Level — the rung the request was served at