OpenAI: GPT-5.5 Pro

Review this 40,000-line TypeScript monorepo codebase I've attached. Identify architectural debt, suggest a migration path to microservices, and flag any security vulnerabilities in the auth layer.

In this illustrative example, the model would produce a structured analysis across three sections: (1) Architectural Debt — identifying tight coupling between the payment and user modules, excessive use of singleton patterns in the service layer, and lack of domain boundaries; (2) Migration Path — proposing a phased extraction starting with the notification service, then the analytics pipeline, with specific interface contracts and a rollback strategy; (3) Security Review — flagging a JWT validation bypass in the refresh token handler and recommending rate-limiting on the password reset endpoint. The response would reference specific file paths and line numbers from the uploaded codebase.

This example showcases the 1M+ token context window — the model can ingest an entire large codebase in one prompt and reason across modules. At $180/Mtok output, a 2,000-word response costs ~$0.36, making exhaustive analysis expensive for frequent iteration.

I'm attaching a 90-page RFP document and our company's last three proposal PDFs. Draft a compliant response to Section 4 (Technical Approach) that reuses our proven methodology but addresses their specific compliance requirements.

In this illustrative example, the model would generate a 3,500-word Technical Approach section that mirrors the structure required by the RFP's Section 4 rubric. It would adapt language from the previous proposals — reusing the Agile delivery framework and risk mitigation tables — while inserting new paragraphs addressing the client's FISMA compliance mandates and their required use of FedRAMP-authorized cloud infrastructure. The draft would cite specific RFP page numbers where requirements appear and flag two areas where the company's standard approach conflicts with the client's procurement rules.

This example demonstrates multi-document synthesis across 400+ pages of input. The model can cross-reference requirements and past work without losing thread. The $30/Mtok input cost means processing this 150k-token prompt costs ~$4.50 — viable for high-value proposals, prohibitive for routine document work.

Analyze the three product mockup images I've uploaded. For each screen, suggest UX improvements for mobile accessibility, identify WCAG 2.2 violations, and rewrite the microcopy to match our brand voice guide (attached as PDF).

In this illustrative example, the model would provide screen-by-screen feedback: for the checkout screen, it would note insufficient color contrast (WCAG 1.4.3 failure) on the 'Apply Coupon' button, suggest increasing tap target size for the payment method selector, and rewrite the error message from 'Invalid card' to 'We couldn't process this card — double-check the number and try again' to match the brand's conversational tone. For the profile screen, it would flag missing alt text on avatar images and recommend relocating the 'Delete Account' link to reduce accidental taps. Each suggestion would reference specific brand voice principles from the uploaded guide.

This example highlights multimodal reasoning — the model interprets visual UI elements, applies technical accessibility standards, and adapts copy to brand guidelines simultaneously. The combination of image and file inputs shows the model's flexibility, though visual analysis quality depends on image resolution and UI complexity.

When 1M+ token context justifies the $180/Mtok output cost

A 4-person litigation support team needs to cross-reference 40+ depositions, contracts, and email threads in a single query—total input around 800K tokens. GPT-5.5 Pro's 1.05M context window handles this without chunking or retrieval hacks, and the model returns a 12K-token summary with inline citations in one pass. At $30 input / $180 output per Mtok, that query costs roughly $26.16—expensive, but faster and more accurate than three hours of paralegal time at $85/hr. The break-even is around 15 complex queries per week; below that, use a cheaper model with RAG. Above it, the time savings and reduced error rate make GPT-5.5 Pro the default for discovery workflows where context density matters more than per-query cost.

Why massive context beats RAG for whole-repo architecture changes

A 12-engineer SaaS team is migrating a 600K-token monorepo from REST to GraphQL. They load the entire codebase—controllers, schemas, tests—into GPT-5.5 Pro and ask it to draft a migration plan with file-level diffs. The model's 1M+ context means it sees every dependency and naming collision without retrieval lag or chunking artifacts. Output cost is high ($180/Mtok), but the team generates 80K tokens of migration code in 90 minutes versus a week of manual mapping. The trade-off: if your refactor touches fewer than 200 files or you're running this daily, a smaller model with vector search is cheaper. For one-time or quarterly architecture rewrites where accuracy and speed justify the spend, GPT-5.5 Pro is the call.

When image + text modality unlocks dense PDF workflows

A 3-person investment research shop ingests 10-Ks, proxy statements, and earnings decks—PDFs with tables, charts, and footnotes spanning 400-600 pages. GPT-5.5 Pro's file and image modalities let them upload the raw PDF, and the model parses both the text and embedded visuals in one pass. The team asks, 'Compare revenue recognition changes across the last three years and flag any footnote discrepancies.' The model returns a 15K-token memo with table references and chart annotations. At $30 input / $180 output per Mtok, a single report costs around $18-24. If you're analyzing fewer than 5 reports per month, use a cheaper OCR + text model. If you're running 20+ deep-dives monthly and need multimodal accuracy, GPT-5.5 Pro pays for itself in reduced manual review time.