Google: Gemini 3.1 Pro Preview (batch)

Analyze this 45-minute earnings call recording and extract: (1) forward guidance changes, (2) management sentiment shifts from Q3, (3) any hedging language around supply chain. Output as structured bullet points.

This example would demonstrate the model processing a multi-modal input (audio file) and returning structured analysis. The model would transcribe the call, identify temporal references to Q3, flag conditional phrases like 'if market conditions permit' or 'subject to supplier availability', and organize findings into three distinct sections. With the 1M+ token context window, the entire 45-minute call fits in a single request, preserving conversational context that shorter-window models would fragment.

Showcases the model's multi-modal strength and massive context capacity — critical for document-heavy workflows. The batch processing mode means this isn't real-time; you'd submit the job and retrieve results later, trading immediacy for cost efficiency at $1/$6 per Mtok versus real-time pricing.

I'm uploading 8 architectural PDFs (site plans, elevations, mechanical drawings) for a mixed-use development. Identify any conflicts between the civil site plan's stormwater routing and the landscape architect's planting zones in sheets C-201 and L-101.

This example would show the model ingesting multiple large PDF files simultaneously, cross-referencing sheet callouts, and identifying spatial conflicts. The output would cite specific drawing coordinates (e.g., 'Grid line D between stations 3+20 and 3+45') where a proposed bioswale overlaps an underground detention system, then suggest whether the landscape or civil design should yield based on code priorities visible in the title blocks.

Highlights file-handling and cross-document reasoning across a 1M+ token budget — eight technical PDFs easily fit. The model can hold all sheets in context simultaneously rather than requiring multi-pass retrieval. Batch mode suits this workflow since design review isn't time-critical and the $1 input cost makes bulk document analysis economical.

Here's a 90-second product demo video. Write a technical FAQ covering the three features shown, anticipate two common objections based on competitor positioning, and draft responses a solutions engineer would give.

This example would demonstrate video comprehension: the model would describe on-screen UI interactions, infer feature benefits from voiceover narration and visual cues, then generate six FAQ entries (three feature explainers, two objection handlers with rebuttals). The responses would match a solutions engineer's tone — specific on capabilities, acknowledging trade-offs where the competitor has an edge, redirecting to differentiated strengths.

Shows video modality and reasoning about competitive context the model must infer (not stated in the video itself). The batch processing constraint means you can't use this interactively during a live demo prep call — you'd queue the job, then retrieve polished FAQ copy 15-30 minutes later at significantly lower cost than synchronous API calls.

When batch processing wins for mixed-media research teams

A 4-person policy shop ingests 200+ documents weekly—PDFs, audio briefings, video testimony, spreadsheets—and needs structured summaries by Friday. Gemini 3.1 Pro Preview handles all five modalities in one call, and the batch endpoint cuts input cost to $1/Mtok (83% cheaper than real-time). The 1M-token window means you can drop an entire hearing transcript plus supporting exhibits into a single request. Run the batch overnight Thursday, review summaries Friday morning. This works if your turnaround is 12+ hours and you're processing 50+ mixed-format files per week. Below that volume, the batch setup overhead outweighs the savings—stick with a real-time multi-modal model and pay the convenience premium.

Batch video analysis for high-volume content platforms

A 12-person creator platform reviews 800 uploaded videos daily for policy violations before publishing. Gemini 3.1 Pro Preview's native video understanding removes the transcription step, and batch pricing makes the economics work at scale—$1 input means analyzing a 10-minute video costs roughly $0.15 in tokens. Queue uploads as they arrive, process the batch every 2 hours, flag violations for human review. The model sees visual context (not just speech), so it catches issues transcripts miss. This setup works when you can tolerate 2-hour review latency and process 500+ videos/day. If you need sub-30-minute moderation or handle under 200 videos daily, real-time endpoints justify the 6x cost increase because batch overhead kills your velocity.

When batch mode handles periodic deep-document work

A 3-person investment fund analyzes 150 earnings reports each quarter—10-Ks, call transcripts, slide decks—extracting risk factors and guidance changes into a comparison table. Gemini 3.1 Pro Preview's 1M-token window fits an entire 10-K plus exhibits in one prompt, and batch pricing ($1 input) makes the quarterly spike affordable. Load all 150 documents Friday after market close, run the batch over the weekend, review extracted data Monday. The multi-modal support handles both text filings and PDF charts without preprocessing. This works for periodic high-volume work (quarterly, annual) where 48-hour turnaround is fine. If you need daily earnings analysis or process under 50 documents per cycle, real-time models cost more per token but save enough analyst time to break even.