OpenAI: GPT-5.4 Mini

Review this Python function for edge cases and suggest defensive improvements. The function merges two sorted lists but occasionally fails in production.

The function assumes both input lists are non-empty and pre-sorted, which creates three failure modes in production. First, passing None instead of an empty list triggers an AttributeError on line 7. Second, unsorted input produces incorrect output without warning—consider adding an optional validation flag. Third, the comparison on line 12 fails when list elements aren't comparable types (mixing strings and integers). Add type hints, validate inputs at the boundary, and consider whether silent failure or an exception better serves your API contract. The merge logic itself is sound for the happy path.

This example shows GPT-5.4 Mini's code review capability within a 400k-token context window—useful for analyzing entire codebases in one prompt. The response balances specificity (line numbers, failure modes) with architectural guidance. However, at $4.50/Mtok output, lengthy reviews of large files become expensive compared to smaller models that handle isolated functions well enough.

Explain how OAuth 2.0 authorization code flow works to a junior developer who understands HTTP but hasn't worked with auth protocols. Include the security rationale for each redirect.

Think of it as a three-party handshake where you never share passwords directly. When a user clicks 'Login with Google' in your app, you redirect them to Google's server with your app's client ID. Google authenticates the user, then redirects back to your app with a temporary authorization code in the URL. Your app's backend exchanges this code (plus a secret key Google gave you during registration) for an access token. The two-redirect design prevents your frontend JavaScript from ever seeing the access token—if an attacker intercepts the first redirect, they get a useless code that expires in seconds and requires your secret to redeem.

Demonstrates the model's ability to translate complex protocols into accessible explanations without sacrificing technical accuracy. The teaching voice is clear and the security framing is appropriate for the audience. The 400k context window matters less here, but the model's training on recent documentation shows in the up-to-date flow description.

I'm attaching a 28-page PDF of our Q3 financial statements. Summarize cash flow trends, flag any line items that moved >15% quarter-over-quarter, and draft three questions for our CFO.

Operating cash flow improved 23% to $4.2M, driven primarily by faster receivables collection (DSO dropped from 47 to 38 days). However, capital expenditures spiked 31% to $1.8M—the footnotes indicate this was planned infrastructure investment, but it compressed free cash flow to $2.4M. Accounts payable increased 19%, suggesting extended payment terms with suppliers. Three questions for your CFO: (1) Is the CapEx spike a one-time investment or the start of a multi-quarter cycle? (2) What's driving the improved receivables collection—process changes or customer mix shift? (3) Are the extended payables strategic or a response to tighter working capital?

Showcases the 400k-token context window handling multi-page financial documents with cross-referencing between sections. The model identifies percentage changes, connects line items to footnotes, and generates contextually relevant questions. The $0.75/Mtok input pricing makes this economical for document analysis compared to manual review, though users should verify numerical accuracy against source data.

When 400K context beats chaining smaller models for legal review

A 12-person legal ops team processes vendor contracts that reference 8-12 prior agreements, policy docs, and compliance frameworks. GPT-5.4 Mini fits the entire contract stack—often 150-200 pages—in a single 400K-token context window, so the model sees cross-references without you building a RAG pipeline or prompt-chaining logic. At $0.75/Mtok input, a 300K-token analysis costs $0.23, roughly half what you'd pay routing the same job through a shorter-context premium model that forces chunking. The trade-off: output is $4.50/Mtok, so if your workflow generates 50K+ token summaries per contract, cost climbs fast. This model wins when input dominates output and you need the full document graph in one pass.

Why Mini's pricing makes it the default for 500+ tickets daily

A 20-seat SaaS support team handles 600 inbound tickets a day, each needing category tagging, urgency scoring, and a two-sentence routing note. Average ticket is 800 tokens in, 150 tokens out. At those volumes, GPT-5.4 Mini costs $0.36 per thousand tickets ($0.0006 input + $0.00068 output), compared to $1.80-$2.40 for flagship models with similar accuracy. The team runs the model in a Zapier-to-Linear pipeline; response quality matches their old manual triage 91% of the time (internal benchmark). The ceiling: if tickets spike past 1,200/day and you need sub-200ms latency, you'll want a dedicated fine-tune or a faster endpoint. Below that threshold, Mini's price and 400K context (for attaching help docs) make it the obvious pick.

When multimodal input and long context justify the output premium

A 4-person e-commerce ops team photographs 80 product SKUs daily—each SKU has 3-5 images showing labels, dimensions, and packaging. They need structured JSON (brand, size, material, compliance marks) extracted into Airtable. GPT-5.4 Mini accepts image + text modalities, so one prompt can ingest all photos for a SKU plus the existing catalog row, then return a single JSON object. Input cost is negligible ($0.75/Mtok for text; image tokens vary but stay low at this scale). The $4.50/Mtok output rate stings if you generate verbose descriptions, but structured extraction averages 600 tokens out, so cost per SKU is ~$0.003. The model's 400K context means you can include the full style guide and 50 example rows as few-shot context without truncation. If your output is under 1K tokens and you need multimodal + long context, this is the play.