Google: Gemini 2.5 Pro (batch)

Analyze this 45-minute earnings call transcript and extract: (1) three material risks mentioned by executives, (2) forward guidance changes vs. prior quarter, (3) analyst sentiment shifts. Format as a structured memo.

This example would produce a structured memo with three sections. The model processes the full transcript (well within its 1M token context window) and identifies material risks like supply chain constraints in APAC, regulatory uncertainty in EU markets, and customer concentration in enterprise segment. The forward guidance section compares Q3 projections against Q2 actuals, noting revenue guidance raised 3-5% while margin expectations compressed. The sentiment analysis flags two analysts shifting from neutral to cautious on near-term growth, citing macro headwinds. The memo maintains executive-level brevity while preserving numerical precision from the source material.

The 1M token context window makes this model practical for long-document analysis that would require chunking in smaller models. Batch processing keeps costs manageable at $0.63/Mtok input for this workload. The output demonstrates structured reasoning across a large input, though real-time use cases would need the standard API.

I'm attaching a 20-second product demo video and three competitor landing pages. Generate a positioning statement that differentiates our workflow automation tool, then draft two headline variations for our homepage hero section.

This example would produce a positioning statement emphasizing the visual workflow builder shown in the video, contrasting it against the code-first approaches visible in competitor screenshots. The model identifies that competitors require YAML configuration or API scripting, while this product offers drag-and-drop canvas editing. First headline variation: 'Automate workflows without writing code—see results in minutes, not sprints.' Second variation: 'Your team's processes, visualized. No developers required.' Both headlines anchor to the video's demonstration of non-technical users building automations, a differentiator the model extracted from multimodal comparison.

Multimodal input (video + images + text) lets the model ground positioning in actual product capabilities rather than generic claims. The batch API suits this use case since positioning work rarely needs sub-second latency. At $5/Mtok output, keeping responses focused (under 200 tokens here) controls costs for creative tasks.

Review this TypeScript monorepo (15 packages, ~80k lines). Identify architectural debt: circular dependencies, duplicated business logic across packages, and API inconsistencies between our REST and GraphQL layers. Prioritize by refactor effort vs. maintenance burden.

This example would produce a prioritized list identifying: (1) High-priority: circular dependency between `@app/auth` and `@app/user-service` causing build-time issues, estimated 3-day refactor to introduce a shared types package. (2) Medium-priority: payment validation logic duplicated in four packages with slight variations, creating bug surface area; consolidation into `@app/payments-core` would take 5 days but reduce test burden by 40%. (3) Lower-priority: REST endpoints return snake_case while GraphQL uses camelCase, affecting 23 client integrations; migration requires coordination but technical work is 2 days. Each item includes file paths, affected LOC counts, and refactor sequencing notes.

The 1M token context window accommodates entire codebases, enabling whole-system analysis that spot-checks miss. Batch processing suits this periodic review workload (weekly or sprint-based) where 24-hour turnaround is acceptable. The model's file and text modalities let it parse both code and documentation, though very large repos may still need selective inclusion to stay under context limits.

When Gemini 2.5 Pro batch beats real-time for research reports

A 5-person market research consultancy writes 8 client reports monthly, each synthesizing 40-60 PDFs, spreadsheets, and slide decks into a 20-page deliverable. Gemini 2.5 Pro batch is the move: the 1M-token context fits entire research packets in one prompt, multimodal file support reads charts and tables without preprocessing, and batch pricing keeps per-report costs under $3 even with 200K-token outputs. The boundary is urgency—batch jobs return in hours, so this works for weekly deliverables but not same-day requests. If your reports need 24-hour turnaround and you're writing 5+/month, batch saves 60% versus real-time models. Below that, the workflow overhead isn't worth it.