Gemini 3.6 Flash vs 3.5 Flash-Lite: $30 vs $8 Monthly API Cost

Decision first: choose Gemini 3.6 Flash when agentic reasoning, coding, and multimodal quality justify a higher successful-task cost. Choose Gemini 3.5 Flash-Lite for extraction, document parsing, subagents, and repeatable high-volume work. Google lists both as generally available. The official rates are $1.50/$7.50 and $0.30/$2.50 per million input/output tokens respectively.

Official pricing snapshot

Model	Input / 1M	Output / 1M	Context	Official role
Gemini 3.6 Flash	$1.50	$7.50	1M	Agentic and multimodal balance
Gemini 3.5 Flash-Lite	$0.30	$2.50	1M	Fast, low-cost high-throughput execution
Gemini 3.5 Flash	$1.50	$9.00	1M	Previous balanced Flash reference

These are official API rates, not relay offers. ComputeUnion keeps channel quotes separate because a lower relay price can reflect different routing, quotas, support, privacy terms, or model mapping.

Monthly cost at two production volumes

The calculation is input millions × input rate plus output millions × output rate.

Workload	Input	Output	3.6 Flash	3.5 Flash-Lite
Production trial	10M	2M	$30	$8
High-volume system	1B	200M	$3,000	$800

Flash-Lite reduces the token bill by about 73% in this fixed workload. That saving only survives if it does not create enough retries, fallbacks, or human corrections to erase the difference.

External capability evidence

Model	Artificial Analysis Intelligence Index	What the score can tell you	What it cannot tell you
Gemini 3.6 Flash (high)	50	Dated independent comparison under the stated configuration	Your task accuracy, latency, or successful-task cost
Gemini 3.5 Flash-Lite	36	Relative general capability in the published evaluation	That every extraction or subagent task needs the larger model

Which model fits which workload?

• Use 3.6 Flash for complex tool use, coding agents, multimodal reasoning, and tasks where failed attempts are expensive.
• Use 3.5 Flash-Lite for classification, extraction, document parsing, routing, simple subagents, and large repeatable queues.
• Route instead of choosing one forever. Start predictable work on Flash-Lite and escalate uncertain or failed tasks to 3.6 Flash.

A practical migration test

1. Replay the same 100–500 representative requests on both models.
2. Measure accepted outputs, retries, latency, input and output tokens.
3. Calculate cost per accepted task, not cost per token alone.
4. Check tool-call compatibility and output schema before moving production traffic.

Clear conclusion

Gemini 3.6 Flash is the safer first choice for difficult agentic and multimodal work. Gemini 3.5 Flash-Lite is the stronger cost choice for high-volume, bounded tasks. The useful architecture is usually a two-tier route: Flash-Lite for the common path and 3.6 Flash for escalation.

Sources and method

Availability, model IDs, context limits, positioning, and prices come from the Google Gemini latest model guide and the Flash-Lite model page. Intelligence Index values come from the dated Artificial Analysis changelog. Monthly costs are deterministic calculations from the official rates; they are not observed customer invoices.