Anthropic: Claude Opus 4.6 (Fast)

Review this Python function for potential race conditions. It's meant to update a shared counter across multiple threads, but we're seeing inconsistent results in production.

The function lacks thread synchronization around the read-modify-write sequence. Between reading the counter value and writing it back, another thread can interleave its own update, causing lost increments. The typical fix is wrapping the entire operation in a threading.Lock() context manager, or switching to threading's built-in atomic counter if you only need increment/decrement. If this counter feeds into financial calculations, consider whether you need stronger consistency guarantees than a simple lock provides — perhaps moving the state into a database with proper transaction isolation.

This example highlights Claude Opus 4.6's ability to diagnose concurrency bugs with architectural context. The 1M-token window means you could paste entire codebases for review, though response latency increases with input size. The model connects the immediate bug to downstream business risk (financial calculations), which is useful but can feel presumptive if your use case is simpler.

I have 47 customer support tickets from last week. Extract common themes, rank them by urgency, and draft a one-paragraph executive summary I can send to our product team.

Three themes dominate: (1) Mobile app crashes on Android 13+ during checkout (12 tickets, high urgency — directly blocks revenue), (2) confusion about the new pricing tier structure (18 tickets, medium urgency — suggests onboarding UX needs work), (3) requests for dark mode (17 tickets, low urgency — quality-of-life feature). Executive summary: 'Last week's support volume spiked around our pricing refresh, with 18 tickets showing users can't distinguish between Pro and Enterprise tiers. More critically, Android 13+ users face checkout crashes (12 reports), which we should prioritize as a revenue blocker. Dark mode requests continue to accumulate but remain non-blocking.'

Demonstrates multi-document synthesis and business prioritization. The model infers urgency from context (revenue impact) rather than requiring explicit labeling. At $150/Mtok output, long summaries across many tickets add up quickly — consider whether you need full drafts or just bullet points for cost control.

Analyze this contract PDF and flag any clauses that deviate from our standard SaaS terms, particularly around liability caps, data retention, and auto-renewal language.

The model would process the uploaded PDF and produce a structured comparison: 'Three deviations found: (1) Liability cap is set at 12 months of fees instead of your standard 3 months (Section 8.2) — significantly increases your exposure. (2) Data retention extends to 5 years post-termination versus your standard 90 days (Section 11.4) — check if this conflicts with your DPA. (3) Auto-renewal notice period is 90 days instead of 30 (Section 2.3) — locks customers in longer, but may reduce churn if that's your goal. The indemnification and IP ownership clauses match your template.'

Showcases file-upload capability and domain-specific reasoning. The model cross-references contract language against implicit 'standard terms' you'd provide in the prompt. However, this is not legal advice — the model can miss subtle clause interactions that a lawyer would catch, so treat output as a first-pass filter, not a compliance guarantee.

When 1M-token context justifies the $150/Mtok output cost

A 4-person litigation support team needs to cross-reference 200+ depositions and contracts in a single query without chunking or retrieval lag. Claude Opus 4.6 Fast handles the entire discovery set in one context window—1M tokens means roughly 750,000 words or 300 dense PDFs loaded simultaneously. The model returns synthesized findings across all documents in under 30 seconds, citing page numbers and exhibit IDs without hallucinating cross-references. At $30 input / $150 output per Mtok, a typical 800k-token load with 5k-token answer costs $27.75 per query. If your team runs fewer than 40 complex discovery queries per month, this beats hiring a contract attorney at $95/hour. Above that threshold, consider a RAG pipeline with a cheaper model.

Fast iteration on board decks when output quality blocks shipping

A 3-person startup finance team writes quarterly board memos that blend narrative, financial tables, and market analysis—documents where a single logic error costs credibility. Claude Opus 4.6 Fast ingests the full data room (cap table, P&L, competitor teardowns) as images and structured files, then drafts sections that reference specific line items without manual lookup. The "Fast" variant returns 2,000-word drafts in 8-12 seconds, enabling real-time editing during partner review calls. At $150/Mtok output, a 2k-word draft costs roughly $0.30—negligible when the alternative is 90 minutes of manual synthesis. If your memos are under 500 words or purely templated, a smaller model at $15/Mtok output saves money without sacrificing speed.

When sub-15-second response time justifies premium pricing

A 12-person SaaS support team handles 300 escalations daily, each requiring analysis of chat history, account metadata, and product docs before routing to engineering or success. Claude Opus 4.6 Fast processes 50k-token context (full conversation thread plus knowledge base excerpts) and returns triage recommendations in under 10 seconds—fast enough to keep the customer on the line. The model's image and file handling means support reps paste screenshots and error logs directly without preprocessing. At $30 input / $150 output per Mtok, a typical 50k-input / 500-token-output triage costs $2.25 per escalation, or $675/day at 300 volume. If your team handles under 100 escalations daily, the speed premium pays off; above that, batch processing with a slower model cuts costs by 60%.