AI Pricing and Access¶
AI pricing and access covers the cost structures, subscription models, and institutional access considerations that researchers face when adopting AI tools.
Context & Background¶
AI tools use various pricing models:
- Free tiers: Limited access to basic models (ChatGPT free, Claude free)
- Pro subscriptions: $20-30/month for enhanced access (ChatGPT Plus, Claude Pro)
- API pricing: Pay-per-token for programmatic access (varies by model and provider)
- Institutional licenses: University-negotiated access (increasingly common)
- Usage-based: Tools like Claude Code with metered billing based on actual usage
Access tier as a research-capability constraint¶
Ortoleva and Sandomirskiy (Markus Academy 166-3) argue the tier you can afford now materially affects what research you can do, and give a recommendation at each price point as of late July 2026: Pro (~$200/mo) — GPT-5.6 Pro in the browser plus Sol Ultra in ChatGPT Work; $20/mo — OpenAI Plus, narrowly over Claude; free — Gemini Pro via Google AI Studio, which handles most basic theory tasks confidently (with the caveat that AI Studio chats train Google's models, which matters for unpublished work).
Their intelligence-versus-stamina framing reframes the budget question: a shallower model that grinds for twenty hours often beats a deeper model run briefly, so the binding constraint is frequently how long you can afford to run something, not how smart it is. Running a top-tier model for an 80-hour proof attempt costs thousands of dollars with an uncertain outcome — a bill academic research accounts are not currently structured to absorb. This is also why the billing mechanism matters as much as the price: a model available under a flat Pro subscription is administratively feasible for a university in a way that metered API access at scale is not.
Practical Implications¶
- Start free, upgrade strategically: Use free tiers to learn, then pay for tools that save you significant time
- Budget for AI tools: $20-200/month is typical for active research use
- Explore institutional access: Check if your university provides AI tool access
- Monitor API costs: Token-based pricing can add up quickly with agentic workflows
- Prefer flat-rate subscriptions for long runs: Subscription-billed agentic products avoid both the cost uncertainty and the procurement friction of API pricing
- Match the model to the problem, not the leaderboard: Ask whether the task rewards depth or persistence before paying for depth
Key Sources¶
- AI-Powered Pipeline to Stress-Test Research Ideas Before PhD Students Spend a Year on Them
- LLM Collaboration and Reasoning -- Generative AI for Economic Research
- Getting Started with Claude Code: A Researcher's Setup Guide
- Getting Started: Claude Code for Economists (Markus Academy Ep. 162-1)
- Vibe Research, or How I Wrote an Academic Paper in Four Days
- Which Model to Use for Theory (Markus Academy 166-3)