Google: Gemini 3.1 Flash Lite

Analyze this quarterly sales spreadsheet and identify the three product categories with declining month-over-month growth. Explain what might be causing each trend.

Looking at your Q3 data, three categories show consistent MoM decline: Electronics (-12% Aug, -8% Sep), Home & Garden (-6% Aug, -9% Sep), and Sporting Goods (-4% both months). Electronics likely reflects seasonal back-to-school saturation—purchases front-loaded in July. Home & Garden's steeper September drop aligns with end-of-summer outdoor project completion. Sporting Goods shows steadier decline, possibly indicating broader discretionary spending pullback rather than seasonality. I'd recommend comparing these patterns against last year's Q3 to separate cyclical from structural issues.

The 1M+ token context window handles large spreadsheets without truncation, making this ideal for financial analysis workflows. The model synthesizes trends across multiple data points and offers plausible business reasoning. At $0.25/Mtok input, processing entire quarterly datasets remains cost-effective, though output quality for nuanced strategic recommendations may require validation against specialized financial models.

I'm attaching a 45-minute lecture video on neural network architectures. Create a structured study guide with timestamps for each major concept and three practice questions per section.

The model would process the video and produce a hierarchical study guide: Section 1 - Perceptrons & Activation Functions (00:00-08:30) covers basic neuron structure, sigmoid/ReLU comparisons, and the XOR problem, with practice questions on activation function selection. Section 2 - Backpropagation Mechanics (08:30-22:15) breaks down gradient descent, chain rule application, and vanishing gradients, including questions on derivative calculations. Section 3 - CNN Architectures (22:15-35:40) details convolution operations, pooling strategies, and ResNet skip connections. Section 4 - Training Strategies (35:40-45:00) addresses batch normalization, dropout, and learning rate scheduling. Each section includes timestamp references and progressively difficult questions.

Multimodal input (video + audio transcription) lets the model extract both visual diagrams and spoken explanations, creating richer study materials than text-only models. The massive context window accommodates full lecture transcripts without chunking. However, the 'Lite' designation suggests this may prioritize speed over depth—complex mathematical derivations in practice questions might lack the rigor of larger Gemini variants.

Review this 80-page technical specification document for our API and generate a migration guide for developers moving from v2 to v3, focusing on breaking changes.

The model would produce a developer-focused migration guide identifying breaking changes across authentication (OAuth 2.0 now mandatory, API keys deprecated), endpoint restructuring (REST paths now follow /v3/resources/{id}/actions pattern), payload format changes (snake_case replaced with camelCase), rate limiting (per-endpoint quotas replace global limits), and error response schemas (now RFC 7807 compliant). Each section includes side-by-side v2/v3 code examples, estimated migration effort, and rollback considerations. The guide highlights that webhook signature verification now requires HMAC-SHA256 instead of SHA1, and batch endpoints have new 100-item limits.

Document understanding across 80 pages demonstrates the practical value of the extended context window for enterprise documentation tasks. At $1.50/Mtok output, generating comprehensive migration guides remains economical compared to manual technical writing. The 'Flash' branding suggests optimized inference speed, making this suitable for iterative documentation workflows where developers need quick answers, though extremely subtle API contract nuances might benefit from human technical writer review.

When Gemini Flash Lite handles 200+ daily tickets under budget

A 4-person SaaS startup gets 200-300 support emails daily and needs to route them to the right person without hiring a coordinator. Gemini Flash Lite is the right call here: at $0.25/Mtok input, you're paying roughly $0.15/day to classify and tag every ticket with context from your help docs. The 1M token window means you can drop your entire knowledge base into each prompt without chunking strategies. Output cost is $1.50/Mtok, but triage responses are short—most tickets generate under 100 tokens of routing metadata. If your ticket volume crosses 1,000/day or you need sub-200ms response times for live chat, step up to the standard Flash model. Below that threshold, Flash Lite keeps your AI budget under $5/month while eliminating the triage backlog.

Flash Lite's multimodal input makes pre-publish review scalable

A 10-person creator network publishes 40-60 short videos weekly and needs to flag potential IP issues, brand-unsafe content, or off-topic submissions before they go live. Flash Lite processes video, audio, and image inputs natively, so you're not paying for separate transcription or frame-extraction pipelines. At $0.25/Mtok input, a 90-second video with audio transcription costs roughly $0.02 to analyze. The model returns a JSON object flagging timestamps and confidence scores for review categories you define. No public benchmarks exist yet for this model, so expect to validate accuracy on your first 50 videos before trusting it unsupervised. If false-positive rates stay under 8%, you've just automated 70% of your pre-publish workload for under $10/week.

When Flash Lite turns 12 weekly calls into structured Notion entries

A 6-person agency runs 12 client calls per week and needs meeting notes, action items, and project updates in Notion within an hour of each call. Flash Lite's audio input support means you upload the recording directly—no Whisper API detour. The 1M token context window handles 90-minute calls with full transcripts and your Notion template in a single prompt. At $0.25 input and $1.50 output per Mtok, each summary costs $0.08-0.12 depending on call length. Output tokens are higher here because you're generating formatted markdown, but you're still under $7/month for the full workload. If your calls involve dense technical jargon or require verbatim quote extraction, test accuracy on 5 calls first—Flash Lite trades some precision for speed and cost. For standard client updates and task tracking, it clears the bar.