OpenAI: GPT-5.4 Mini (batch)

Review this Python function for edge cases. It's supposed to merge overlapping time intervals but users report crashes on empty input and single-element lists.

The function assumes `intervals` has at least two elements—line 8 will throw IndexError if len(intervals) < 2. Add a guard: `if len(intervals) <= 1: return intervals` at the top. Also, the sort at line 3 mutates the input list; consider `sorted(intervals, key=lambda x: x[0])` to avoid side effects. The merge logic itself (lines 10-14) looks sound for overlapping ranges, but you'll want a test for adjacent non-overlapping intervals like [(1,2), (3,4)] to confirm they stay separate.

This example highlights GPT-5.4 Mini's solid code-review instincts within a 400k-token context window—useful for pasting entire modules. The batch API's lower cost ($0.38/Mtok input) makes it viable for nightly CI scans, though the 24-hour turnaround means it won't catch bugs in real-time PR reviews.

Summarise the last six months of meeting notes in this 80-page document. Focus on recurring action items that never got resolved and any budget concerns mentioned more than once.

Across 22 meetings, three action items appear unresolved: (1) migrating the analytics pipeline to the new vendor (mentioned in Jan, Mar, May notes but no closure), (2) hiring a second DevOps engineer (discussed in Feb, Apr, Jun—still 'in progress'), (3) finalising the Q3 marketing budget (flagged in Apr and Jun as blocked by finance approval). Budget concerns: the cloud-infra line item exceeded forecast in four consecutive months (Mar–Jun), and the June 14 note flags a potential 18% overspend if the trend continues into Q3.

The 400k-token window handles the full 80-page doc in one pass, so no chunking or re-prompting needed. GPT-5.4 Mini's summarisation is accurate but occasionally generic—'in progress' and 'blocked' appear verbatim from the source rather than synthesising why these items stalled. For $0.38/Mtok input, batch mode is cost-effective for monthly or quarterly doc reviews where same-day turnaround isn't critical.

Generate five subject lines for a product launch email. The product is a B2B SaaS tool that automates invoice reconciliation. Target audience: finance directors at mid-market companies. Tone: professional, benefit-focused, no hype.

1. Reconcile invoices in minutes, not days 2. Automate your month-end close—starting this quarter 3. Cut reconciliation errors by 80% with automated matching 4. Finance teams save 12 hours/week on invoice workflows 5. Close your books faster: see how automated reconciliation works Each line leads with a tangible outcome (time saved, error reduction, faster close) rather than feature jargon. Line 3's '80%' claim should be backed by a case study in the email body; line 4's '12 hours/week' assumes you have usage data to support it.

This example shows GPT-5.4 Mini's competence at constrained creative tasks—the subject lines stay on-brief and avoid marketing clichés. The model's self-awareness (flagging the need for data to back claims) is a plus, though the output is safe rather than inventive. At $2.25/Mtok output, batch mode suits bulk content generation (e.g. 50 subject-line variants for A/B testing) where you can queue requests overnight and review results the next morning.

Why batch mode wins for high-volume overnight document jobs

A 12-person legal ops team processes 800-1200 vendor contracts nightly, extracting clauses and flagging non-standard terms into their CRM. GPT-5.4 Mini in batch mode cuts their per-document cost to roughly one-third of real-time pricing—at $0.38/Mtok input, a 15k-token contract costs under a cent to process. The 400k context window handles even merged multi-party agreements without chunking. Batch jobs complete within 24 hours, which fits their workflow: contracts arrive by 6pm, results populate by 8am. If you need sub-hour turnaround or interactive clarification loops, real-time GPT-5.4 Mini is the better call. For overnight pipelines where cost per document matters more than latency, batch mode is the obvious choice.

When to use batch Mini for e-commerce catalog captioning at scale

A 9-person DTC brand uploads 2000-3000 product photos monthly and needs SEO-friendly captions with alt-text for each. GPT-5.4 Mini batch handles image+text input at $0.38/Mtok, making per-image costs negligible—most captions generate under 500 tokens output, so 3000 images run about $3.40 total. The team queues photos Friday afternoon and reviews captions Monday morning. Batch latency (up to 24 hours) is fine because they're not blocking a live storefront update. If you're captioning user-uploaded images in real-time or need captions within minutes for same-day launches, real-time Mini or a vision-specialized model makes more sense. For monthly catalog refreshes where you can wait overnight, batch Mini delivers the lowest per-image cost.

How batch Mini handles low-frequency, high-volume text analysis

A 5-person HR consultancy runs employee engagement surveys for 40 clients quarterly, analyzing 8000-12000 open-ended responses per cycle. They paste all responses into a single prompt (the 400k window fits ~300k tokens of survey text), ask for sentiment scoring and theme extraction, then export results to their reporting tool. At $0.38 input and $2.25 output per Mtok, a 300k-token input with 20k-token output costs about $0.16 total—orders of magnitude cheaper than real-time calls. Batch latency doesn't matter because they schedule analysis 48 hours before client presentations. If you're analyzing surveys weekly or need same-day insights, real-time Mini is worth the premium. For quarterly cycles where cost per survey matters and you can plan around 24-hour delivery, batch Mini is the clear winner.