Agentic search and document processing
Good first candidate when request volume, latency, and unit cost dominate.
Test retrieval quality, structured output validity, and retry rate.
Google model field guide · checked July 29, 2026
Flash-Lite is a throughput decision, not a universal downgrade.
Gemini 3.5 Flash-Lite targets high-volume, latency-sensitive production traffic. The useful question is whether it meets your acceptance bar with fewer total resources—not whether its token price is the lowest number on a page.
Input at launch
$0.30 / MTok
Output at launch
$2.50 / MTok
Flowith status
Verify live
Where Flash-Lite earns a pilot
Coding subagents
Use configurable thinking levels to trade speed for multi-step work.
Measure accepted patches and reviewer minutes, not code volume.
Computer use
Google lists computer use as a built-in tool.
Validate permissions, target-site behavior, recovery, and human confirmation.
High-stakes decisions
Low cost does not remove review requirements.
Use domain checks, refusal handling, logging, and escalation.
Evaluation
Measure the whole job
Build a representative set, then record p50 and p95 latency, output tokens, tool calls, retries, schema failures, accepted outputs, and reviewer minutes. Repeat the same set on 3.6 Flash before selecting a default.
Thinking levels are a workload control
Google says minimal and low thinking can prioritize speed and cost, while higher levels can support multi-step subagent work. Treat the level as part of the evaluated configuration and re-test after provider changes.
Gemini 3.5 Flash-Lite FAQ
What is Gemini 3.5 Flash-Lite?
Gemini 3.5 Flash-Lite is Google's July 2026 high-throughput Flash model for low-latency and cost-sensitive agentic work.
What was the launch price for Gemini 3.5 Flash-Lite?
Google published $0.30 per million input tokens and $2.50 per million output tokens. Verify the live API pricing page because provider pricing can change.
Is Flash-Lite always faster or cheaper for a complete task?
No. Provider rates and speed are inputs. Retries, longer outputs, tool loops, failures, and human review determine completed-task cost.
Is Gemini 3.5 Flash-Lite available in Flowith?
This page does not assert Flowith availability. Check the live workspace model selector for the current answer.