Anthropic: Claude Haiku 4.5

Review this Python function for edge cases. It's supposed to parse ISO dates but users report crashes on malformed input.

The function lacks input validation before calling `datetime.fromisoformat()`. If the string doesn't match ISO 8601 format, it raises `ValueError` without handling. Add a try-except block around the parse call, return `None` or a sentinel for invalid input, and consider logging the malformed string for debugging. Also check for empty strings before parsing—`fromisoformat('')` will fail. The current implementation assumes well-formed input, which breaks in production when users paste arbitrary text.

Haiku 4.5 excels at practical code review with a 200K token context window, letting you paste entire modules for analysis. The response is direct and actionable. Trade-off: at $5/Mtok output, verbose explanations cost more than terser models, though input pricing remains competitive at $1/Mtok.

Summarise the key risks and mitigation strategies from this 40-page vendor security questionnaire. Focus on data residency and access controls.

**Data Residency Risks:** Vendor stores EU customer data in US-based AWS regions (pages 12-14), creating GDPR transfer complications. Mitigation: they offer EU-only hosting as an enterprise add-on. **Access Controls:** Role-based access uses custom implementation rather than industry-standard protocols (page 23). No MFA enforcement by default. Mitigation: they commit to SOC 2 Type II audit completion by Q3. **Remaining gaps:** No data deletion guarantees under 90 days; subprocessor list updated quarterly, not real-time.

The 200K context window handles long compliance documents without chunking. Haiku 4.5 extracts structured insights quickly, making it suitable for legal and procurement workflows. Trade-off: nuanced contract interpretation may still require a larger model for high-stakes decisions.

Generate five subject lines for a product launch email. The product is a calendar app for remote teams. Tone: professional but warm.

1. Your team's calendar just got smarter 2. Finally—scheduling that doesn't feel like work 3. We built the calendar remote teams actually use 4. Introducing [Product Name]: calendars, minus the chaos 5. See everyone's availability without the email tennis Each avoids hype while emphasising the core benefit (reduced coordination friction). Option 3 uses social proof framing; option 5 highlights a specific pain point. Test options 2 and 5 for highest open rates based on your audience's communication style.

Haiku 4.5 handles creative tasks with practical framing—useful for marketing teams who need quick iterations. The multimodal capability means you could include brand guidelines as images. Trade-off: output lacks the stylistic flair of models trained specifically for creative writing.

When Haiku 4.5 handles 500+ tickets daily under budget

A 12-person SaaS company routes 600 inbound support emails daily through Haiku 4.5 for intent classification and urgency scoring before human review. At $1/Mtok input, the 200K context window lets you batch 40-50 tickets per call without chunking, cutting API overhead by 80% compared to smaller-window models. Output cost ($5/Mtok) stays negligible because classification responses are under 100 tokens. The model runs stable on structured tasks—tagging, routing, sentiment—where you don't need frontier reasoning but can't afford GPT-4 class pricing at scale. If your tickets average under 2K tokens and you're processing 300+ daily, Haiku 4.5 pays for itself in week one. Beyond 1,000 tickets/day, consider caching strategies or batch endpoints to push cost efficiency further.

Why Haiku 4.5 fits small dev teams documenting legacy codebases

A 4-engineer startup inherits 80K lines of undocumented Python across 200 files and needs inline docstrings plus a README before the next funding round. Haiku 4.5's 200K window accepts 15-20 files per prompt, letting the team process the entire repo in 12 calls instead of 200. Input pricing at $1/Mtok makes the full job cost under $8, and the model handles straightforward summarization without hallucinating function signatures. It won't architect new systems or debug complex logic—frontier models win there—but for mechanical documentation where context breadth beats reasoning depth, Haiku 4.5 delivers at one-tenth the cost of Opus or GPT-4. If your codebase exceeds 500 files or requires architectural analysis, budget for a hybrid approach with a stronger model on critical modules.

How Haiku 4.5 processes 200 invoices nightly for accounting teams

A 3-person accounting firm scans 200 vendor invoices monthly (mix of PDFs and images) and extracts line items, totals, and tax fields into a spreadsheet. Haiku 4.5's image modality handles the scans directly, and the 200K context window lets them stack 8-10 invoices per API call, reducing round-trips and latency. At $1 input per Mtok, the monthly batch costs under $3 even with image tokens priced higher than text. The model's structured output reliability—critical for financial data—sits between GPT-3.5 and GPT-4, accurate enough for human spot-checks but not for zero-touch reconciliation. If your invoice volume exceeds 500/month or error tolerance drops below 2%, test against GPT-4o mini for the accuracy-cost trade-off. For this firm's 200-invoice, human-reviewed workflow, Haiku 4.5 is the floor price that works.