I am starting a master’s in Advanced SoftwareComputers Electronics and Technology

Task prompt

I am starting a master’s in Advanced Software Engineering at King’s College London and expect to build AI portfolio and coursework systems. I want a reusable cost-control and evaluation system for text-generation APIs rather than choosing models from headline prices or generic leaderboards. Compare OpenAI API, Anthropic Claude API and Google Gemini API. Select exactly two current text models per provider: one cost-optimised option and one higher-capability option, each with publicly stated API pricing and suitability for text or structured-output work. Use current official pricing, model, token-accounting, caching, batch, rate-limit and usage-monitoring documentation. Record unavailable information as “not stated” and do not treat differently named billing units as equivalent without evidence. Model three fixed monthly workloads: structured extraction at 2,000 uncached input tokens and 300 output tokens for 10,000 requests; a RAG assistant at 12,000 input tokens and 1,000 output tokens for 2,000 requests, with 60% of input treated as cacheable only where official rules support it; and agentic coding at 20 calls per session, 6,000 input and 1,500 output tokens per call, for 100 sessions with a 10% retry rate. Use £50 as the development budget and £250 as the pilot budget, but make both editable. Build a provider-neutral calculator, an 18-case evaluation pack and a routing policy that can choose different models for different workloads. Produce `ai_api_cost_evaluation_system.md`, `ai_api_cost_calculator.csv` and `ai_api_evaluation_cases.csv`. Include exact source URLs, access dates, formulas, assumptions, conflicts and “not stated” fields so the system can be refreshed when prices or models change. Do not create accounts or API keys, sign in, run paid calls, change spending limits or transmit prompts. This is a researched planning and evaluation design task, not a purchase or live benchmark.