Project Usage & Subscription Cost Analysis

Oriol Vallès Codina · April 2026

Overview

This document summarises AI and software subscription costs (March–April 2026), how they map to different research projects, and the key tools used in each workflow.

Subscriptions: What Each Service Does

Claude (Anthropic)

Claude Pro / Claude Code is the primary AI assistant and agentic coding environment used across all research projects. In this workflow it is used as a pair programmer and research assistant: it reads and writes code (R, Python, JavaScript), navigates codebases, builds data pipelines, edits HTML/CSS/JS for interactive outputs, debugs errors, and helps draft and edit academic writing. Claude Code (the CLI tool used here) goes beyond chat — it executes shell commands, edits files, runs scripts, and manages git. The Claude Sonnet 4.6 model is the default; Opus 4.7 is available for complex reasoning tasks.

ChatGPT Plus (OpenAI)

ChatGPT is used as a secondary AI assistant, primarily for long-form writing, mathematical derivations, and cases where a second AI perspective is useful for cross-checking. GPT-4o handles most tasks; the o1/o3 reasoning models are used for hard mathematical or logical problems (equation notation, formal model setup).

Windsurf Pro (Codeium / Exafunction)

Windsurf is an AI-powered IDE (code editor), similar to VS Code but with deep in-editor AI assistance. It provides inline code completion, refactoring suggestions, and a chat interface directly in the editor. Used primarily for Python and R development where IDE-level autocomplete is valuable. Complements Claude Code by operating at the file-editing layer while Claude Code handles multi-file projects and shell operations.

Posit Connect Cloud

Posit Connect Cloud is the hosting platform for R-based interactive applications (Shiny apps). It was used to host the CVCE Shiny app (the original R Shiny version of the Clean Value Chain Explorer) so that collaborators at NZIPL/Johns Hopkins could access it via a web URL. Cancelled in March 2026 when the project moved to a pure static-HTML architecture deployed on GitHub Pages (no server needed).

Receipts Summary: March 6 – April 30, 2026

By subscription (USD, EUR@1.17 for Anthropic)

EUR→USD conversion: 1.17 (Wise transfer calculator rate, April 2026)

Anthropic detail (invoice sequence NJJAFZLE-XXXX)

April 2026 alone: €279.50 (8 charges: 1 Pro sub + 7 overages of €6–€46 each)

Project Attribution (speculative, based on git history + file timestamps)

Timeline narrative

• Mar 6–22: Claude Pro starts. CVCE Phase II setup — workplan render scripts, dendrite snapshots, deploy pipeline. Usage within Pro limits (no overages).
• Mar 23 – Apr 3 (LEEDS_MODEL sprint): Intensive MRIO equation work for Ecological Economics submission. Two €5 top-ups in March, then €20 × 3 in a 2-day submission sprint (Apr 1–2). Paper submitted to Ecological Economics SI on April 3.
• Apr 8 (CVCE resumes): 13 commits in one day — PC scatter viewer, type→role rename across entire codebase, OneDrive sync. First large top-up (€38.25).
• Apr 10–24 (CVCE intensive): Global Atlas v1, Chile Atlas, Brazil Atlas, Intelligence Report initial build. Second large top-up Apr 15.
• Apr 22+ (CONSERVATION parallel): Conservation/thermodynamics papers start appearing alongside CVCE. Apr 25: sraffa_acs_PRE, value_conservation_paper; Apr 29: econ_thermo_manuscript; Apr 30: full manuscript. Top-ups Apr 28–29 cover both CVCE (IR map debugging) and CONSERVATION (renders).

Cost Efficiency Assessment

April was an outlier

April 2026 was an exceptionally intensive month: LEEDS_MODEL submission + CVCE Phase II acceleration + CONSERVATION parallel work. Normal monthly spend should be lower.

Claude Max recommendation

At current usage intensity, Claude Max ($100/month) is cost-efficient:

• April Anthropic spend: €279.50 (~$327) — 3.3× what Max would cost
• Break-even: ~5 extra usage top-ups/month (you hit 7 in April alone)

With NZIPL Team subscription

If Johns Hopkins NZIPL provides a Claude Team seat:

• CVCE work (~51% of usage) moves to lab account
• Personal usage (CONSERVATION + other personal research) ~€120–130/month
• Personal Max ($100) still likely worth it given CONSERVATION intensity
• Key question: Can the Team seat be your personal account? If yes, Max may be redundant during CVCE-heavy periods.

Productivity Impact of Switching to Claude Code

Claude Code (Anthropic’s agentic CLI) was adopted on March 6, 2026, replacing a multi-tool stack (ChatGPT + Windsurf + Posit). The productivity impact, measured by git commits per week across all active repositories, was immediate and substantial.

Commits per week: before vs. after Claude Code

Before: 28-week lab period (211 total commits). After: 8-week Claude Code period (352 commits).

Monthly CVCE commit trajectory

What the numbers understate

• CONSERVATION papers (active Apr 22–30, multiple files modified daily) have no git repository — this parallel workstream is not captured in commit counts.
• Tool consolidation: ChatGPT Plus cancelled Apr 29, 2026. Windsurf ended Mar 2026. Posit ended Feb 2026. One subscription replaced four, at lower combined cost during intensive periods.
• Quality dimension: commit frequency measures output volume, not quality. The LEEDS_MODEL paper was submitted to a peer-reviewed journal (Ecological Economics SI) during this period; the CVCE Intelligence Report went from concept to deployed interactive HTML in under two weeks.

Files

• docs/receipts/subscription_costs.csv — full ledger (41 rows + total)
• docs/receipts/pdf/ — 29 PDFs named {Service}_{YYYY-MM-DD}.pdf
• docs/receipts/Project_Usage.pdf — lab co-directors version (rendered PDF)

Generated May 1, 2026. EUR/USD conversion: 1.17 (Wise)