How much does Gemini 3.5 Flash-Lite cost?

As of July 22, 2026: $0.30/MTok input and $2.50/MTok output. That's slightly higher than the previous Gemini 3.1 Flash Lite ($0.25/$1.50) — Google traded a small price hike for material quality improvement. Still 5x cheaper on input and 3x cheaper on output than Gemini 3.6 Flash ($1.50/$7.50). Compared to competitors: DeepSeek V4-Flash off-peak at $0.14/$0.28 is still 2x cheaper on input and ~9x cheaper on output. Kimi K3 at $3/$15 is 10x more expensive on input and 6x more expensive on output. GPT-5.6 Luna at $3/$12 is 10x more expensive on input and 4.8x more expensive on output. Google is competing on 'cheap enough for volume, Google Cloud integrated' rather than absolute cheapest.

When should I use Gemini 3.5 Flash-Lite instead of 3.6 Flash?

Use Gemini 3.5 Flash-Lite when volume and latency matter more than best-in-class reasoning. Sweet spots: real-time chat interfaces where sub-500ms first-token latency matters, classification and extraction at millions of requests per day, autocomplete and inline suggestions, batch pipelines for log analysis or content moderation, multi-tenant SaaS where per-request cost dominates margins, and any workload where you'd otherwise evaluate DeepSeek V4-Flash for cost but need Google Cloud integration or Google's data governance guarantees. Use 3.6 Flash instead for: agentic coding with tool chains (Flash-Lite may loop more due to weaker reasoning), long-context synthesis, complex multi-step reasoning, and full multimodal video/audio workloads.

How does Gemini 3.5 Flash-Lite compare to DeepSeek V4-Flash?

DeepSeek V4-Flash off-peak at $0.28/MTok output is ~9x cheaper than Flash-Lite's $2.50/MTok. On aggregate benchmarks the two are roughly comparable at mid-tier standard tasks — both trail GPT-5.6 Luna and Gemini 3.6 Flash on hardest reasoning but handle 80-90% of production workloads well. Flash-Lite advantages over V4-Flash: no geopolitical concerns (V4 is Chinese-origin), tighter Google Cloud integration (Vertex AI, Workspace, BigQuery), lower first-token latency in most regions, no peak-hour pricing gotcha (V4 doubles cost 1-4 AM UTC and 6-10 AM UTC). V4-Flash advantages over Flash-Lite: 9x cheaper on output tokens, MIT-family open weights (self-hosting possible), no data residency concerns for teams already comfortable with DeepSeek. Router pattern: Flash-Lite for US/EU enterprise and Google Cloud native pipelines; V4-Flash for cost-dominated global batch.

Quick Answer

What Is Gemini 3.5 Flash-Lite? Google's Cheap Tier (July 2026)

Published: July 22, 2026

What Is Gemini 3.5 Flash-Lite? Google’s Cheap Tier (July 2026)

Gemini 3.5 Flash-Lite is Google’s cheap, high-throughput Gemini variant launched July 21, 2026 as part of a three-model release (alongside 3.6 Flash and 3.5 Flash Cyber). Priced at $0.30/MTok input and $2.50/MTok output, it’s Google’s response to DeepSeek V4-Flash and Kimi K3 dominating the cost-sensitive segment of the mid-tier market.

If you looked at Gemini 3.5 Flash and thought “too expensive for what I need,” this is Google’s answer.

Last verified: July 22, 2026

The Fast Facts

Attribute	Value
Released	July 21, 2026
Positioning	Cheap, fast, high-throughput
Input price	$0.30/MTok
Output price	$2.50/MTok
Context window	1M input tokens
Multimodal	Text + image (limited video/audio)
SWE-Bench Pro	54.2% (beats Gemini 3 Flash’s 49.6%)
License	Proprietary (Google Cloud terms)
Available in	Gemini API, AI Studio, Vertex AI

Why Google Launched It

Two things forced Google’s hand on a cheap tier:

1. DeepSeek V4-Flash’s July 19, 2026 GA launch at $0.28/MTok output established that frontier-adjacent capability at ultra-low prices is feasible. Prior to V4, Google’s cheapest tier could sit at ~$1.50/MTok output without competitive pressure. Now sub-$0.50 is the new floor.

2. Enterprise AI budgets are cost-constrained. Info-Tech Research Group reports 79% of enterprises experienced AI cost overruns in the past 12 months. CIOs pushed back on “let’s just use Gemini 3.5 Flash” defaults because at $9/MTok output, high-volume workloads got expensive fast. Google needed a “safe” (US company, Google Cloud native) cheap alternative or watch traffic migrate to DeepSeek.

Gemini 3.5 Flash-Lite is Google saying: “You want cheap? Here’s cheap. Still Google Cloud.”

Pricing Comparison

Model	Input $/MTok	Output $/MTok	Positioning
DeepSeek V4-Flash off-peak	$0.14	$0.28	Cheapest, Chinese-origin
Gemini 3.5 Flash-Lite	$0.30	$2.50	Cheap + Google Cloud
DeepSeek V4-Flash peak	$0.28	$0.56	Cheapest peak (1-4, 6-10 UTC)
Gemini 3.6 Flash	$1.50	$7.50	Default workhorse
GPT-5.6 Luna	$3.00	$12.00	Best mid-tier capability
Kimi K3	$3.00	$15.00	Best benchmarks open-weight

Practical placement: Flash-Lite sits between DeepSeek V4-Flash (cheaper) and Gemini 3.6 Flash (more capable). It’s not the cheapest — but it’s the cheapest option that doesn’t come with geopolitical or self-hosting complexity.

Note on Price Vs Previous Flash Lite

Ars Technica reported the new 3.5 Flash-Lite is slightly more expensive than the previous 3.1 Flash Lite ($0.25/$1.50 → $0.30/$2.50). Google’s message: “small price hike, big capability jump.” If you’re on 3.1 Flash Lite and were happy, evaluate whether the capability improvement is worth the ~40% price increase for your workload.

What Flash-Lite Is Good At

1. Real-Time Chat Interfaces

Sub-500ms first-token latency in most regions. For chatbots, support interfaces, and interactive assistants where perceived response time drives user satisfaction, Flash-Lite is fast enough that human users don’t feel latency.

2. High-Volume Classification and Extraction

Millions of short requests per day. Content moderation, sentiment analysis, entity extraction from documents, log parsing, PII detection — all these workloads process short inputs and generate short outputs at massive volume. Flash-Lite’s low per-request cost dominates.

3. Autocomplete and Inline Suggestions

Very short outputs, very high frequency. Text editors, IDE autocomplete, search suggest, form auto-fill. Flash-Lite’s low output-token cost matters more than absolute quality for these use cases — users see many suggestions, accept one, move on.

4. Batch Pipelines

Overnight or weekly bulk processing. Log analysis, transcript processing, email categorization at scale, RAG embedding backfills that need cheap generation. If you can tolerate batch latency, Flash-Lite at $2.50/MTok output beats most alternatives that offer Google Cloud integration.

5. Multi-Tenant SaaS With Cost-Constrained Margins

Product teams shipping “AI features” in $10-30/mo subscription products need per-request cost measured in cents, not dollars. Flash-Lite is the tier where “unlimited AI usage” plans become viable.

Where Flash-Lite Is NOT the Answer

Skip Flash-Lite for:

Agentic coding with tool chains — smaller model may loop more, canceling out per-token savings. Use 3.6 Flash.
Complex multi-step reasoning — Flash-Lite trails 3.6 Flash and Luna materially. Use those for hardest tasks.
Long-context synthesis (>200K tokens) — Flash-Lite has 1M context but weaker at deep synthesis over long context. 3.6 Flash is safer for book-length or codebase-scale analysis.
Full multimodal video/audio — check API docs; 3.6 Flash is the safer choice for video and audio understanding.
Cheapest-possible batch — DeepSeek V4-Flash off-peak wins by 9x if you can accept Chinese-origin infrastructure.

Router Pattern With Flash-Lite

Recommended production router for $10K+/mo AI spend:

def route_request(request):
    # Cheap tier: high volume, simple
    if request.is_classification_or_extraction() or \
       request.is_autocomplete() or \
       request.expected_output_tokens() < 500:
        return gemini_3_5_flash_lite

    # Multimodal or long-context
    if request.needs_video_audio() or request.needs_long_context():
        return gemini_3_6_flash

    # Hardest reasoning
    if request.is_hardest_task():
        return gpt_5_6_luna  # or Sol if budget allows

    # Default agentic workload
    return gemini_3_6_flash

Expected outcome: 40-60% cost reduction vs “just use 3.6 Flash” defaults if your traffic mix skews toward classification and short-response workloads.

The Strategic Signal

Flash-Lite completing Google’s Flash tier segmentation (Lite → Standard → Cyber) tells us three things about Google’s 2026-2027 strategy:

1. Google accepts it can’t beat frontier on capability right now. 3.5 Pro is delayed; Gemini 4 pretraining is teased. Google is doubling down on Flash tier segmentation as the near-term battleground.

2. Google is copying DeepSeek’s pricing playbook. DeepSeek pioneered peak-valley + Pro/Flash segmentation. Google now has cheap (Lite) / standard (3.6 Flash) / gated-specialty (Cyber) — mirroring the segmentation approach without the peak-valley pricing.

3. Frozen v2 (2028) is the endgame. Google’s custom silicon should deliver 6-10x efficiency over current TPUs. If those savings pass through to API pricing, Gemini’s cheap tier could match V4-Flash’s sub-$0.50/MTok output by 2029.

Bottom Line

Use Gemini 3.5 Flash-Lite if:

You have high-volume, cost-sensitive workloads (millions of requests/day).
You need Google Cloud integration (Vertex AI, Workspace, BigQuery).
You can’t or won’t use Chinese-origin models (V4-Flash) for compliance, geopolitical, or preference reasons.
Your typical request is short-input, short-output.

Use 3.6 Flash instead if:

Reasoning quality matters more than per-request cost.
You need agentic tool chains, long-context synthesis, or complex multimodal.

Use DeepSeek V4-Flash instead if:

Absolute lowest cost is dominant.
You can accept Chinese-origin infrastructure.
You want the option of self-hosting via MIT-family open weights.

Sources

Google DeepMind announcement: blog.google - Gemini 3.6 Flash, 3.5 Flash-Lite, 3.5 Flash Cyber
Ars Technica coverage: Google reveals faster and cheaper Gemini 3.6 Flash
TechCrunch launch coverage: Google releases three new Gemini models — but no 3.5 Pro
MarkTechPost analysis: Google Releases Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber