Google: Gemini 3.5 Flash Lite

Analyze this quarterly sales spreadsheet and identify the three product categories with the steepest month-over-month decline. Explain what might be causing each drop.

Looking at your Q3 data, Electronics saw a 23% MoM decline in September, likely due to back-to-school season ending and customers waiting for holiday releases. Home & Garden dropped 18%, typical for late summer as outdoor spending shifts indoors. Apparel fell 15%, which tracks with the gap between summer clearance and fall collection launches. All three categories show historical seasonal patterns rather than concerning structural issues—your YoY comparisons remain positive across the board.

Demonstrates multimodal file analysis with the 1M+ token context window handling full spreadsheets without summarization. The model connects data patterns to business context effectively. At $2.50/Mtok output, lengthy analytical responses cost more than competitors, so concise prompting matters for budget-conscious teams.

I'm attaching a 45-minute recorded team meeting (video). Pull out the three action items assigned to Sarah, with timestamps and who assigned them.

Sarah received three assignments: (1) Draft the API migration plan by Friday—assigned by Marcus at 12:40. (2) Schedule follow-up calls with the design team about mobile layouts—assigned by Jennifer at 28:15. (3) Review the security audit findings and flag any blockers—assigned by Marcus again at 41:50. All three came with explicit deadlines mentioned in context.

Video understanding across the full context window lets the model process long meetings without chunking. The $0.30/Mtok input pricing makes this economical for regular meeting analysis. However, with no public benchmarks available, accuracy on complex video reasoning tasks remains unverified—test on your actual meeting formats first.

Convert this technical architecture diagram (image attached) into a Mermaid flowchart with proper node labels and relationship annotations.

```mermaid graph TD A[Client App] -->|HTTPS| B[API Gateway] B -->|Auth Check| C[Auth Service] B -->|Route| D[User Service] B -->|Route| E[Payment Service] D -->|Query| F[(User DB)] E -->|Query| G[(Transaction DB)] E -->|Webhook| H[Stripe API] ``` I've preserved your three-tier structure with the authentication middleware layer. The payment service's external Stripe dependency is clearly marked as a webhook relationship.

Shows practical image-to-code conversion with attention to architectural semantics, not just visual layout. The 'Lite' designation suggests optimizations for speed over maximum accuracy—complex diagrams with overlapping elements or ambiguous arrows may need prompt refinement. Strong value at this price point for routine documentation tasks.

When Flash Lite wins on high-volume visual classification

A 4-person e-commerce team needs to tag 200,000 product photos monthly with category, color, and style attributes. Gemini 3.5 Flash Lite handles this at $0.30/Mtok input—roughly $6 per 100k images if each prompt averages 200 tokens. The 1M-token context window means you can batch 50+ images per call to amortize overhead. Output cost ($2.50/Mtok) stays low because tags are short structured responses. The trade-off: no public benchmarks exist yet, so you're betting on Google's track record with Flash models. If accuracy drops below 92% on your validation set, step up to the standard Flash tier. For teams processing visual data at scale where cost per unit matters more than cutting-edge accuracy, Flash Lite is the right starting point.

Flash Lite for real-time call summaries under budget pressure

A 12-person SaaS support team closes 300 tickets daily and needs one-paragraph summaries written within seconds of each call ending. Gemini 3.5 Flash Lite processes audio natively, so you skip transcription APIs and their latency. At $0.30 input per Mtok, a 20-minute call (roughly 3,000 tokens of audio context) costs under a cent to summarize. The model's multimodal design means you can also ingest screenshots customers share during calls without stitching tools. The boundary: if your team needs sentiment scoring or escalation prediction beyond plain summaries, you'll want a model with published benchmark data on classification tasks. For straightforward summarization at high frequency, Flash Lite delivers the speed and cost structure small teams need.

When Flash Lite handles video review for community platforms

A 6-person creator platform reviews 1,500 user-uploaded videos weekly for policy violations before publishing. Gemini 3.5 Flash Lite ingests video directly, so you avoid frame-extraction preprocessing that burns engineering time. The 1M-token context window fits 10+ minutes of video content per call, and at $0.30/Mtok input, a 5-minute video costs roughly $0.02 to scan for nudity, violence, or spam. Output tokens are minimal—just a binary flag and timestamp—so the $2.50/Mtok output rate barely registers. The risk: without public benchmarks, you need to validate false-negative rates on your own content before going live. If your moderation SLA requires 99%+ recall, budget time for a labeled test set. For platforms under 10k videos/month where cost and native video support outweigh benchmark transparency, Flash Lite is the pragmatic choice.