Migration reports

LLM migration decisions, fully worked.

Each report takes one realistic workload and answers a single question — should this team switch, and to what? — by costing every credible option against the same verified pricing data, then weighing the engineering effort each switch actually requires. These are free, complete samples of the analysis the Migration ROI Simulator produces on your own numbers.

90-second demo — token counter → workload simulator → a report exactly like the ones below.

GO — but stay on OpenAI
A Series-A RAG startup is paying $360k/yr for GPT-5.5. Should they switch?
A worked decision memo: prompt caching, an in-family downgrade to GPT-5.4, and a cross-provider move to Claude Sonnet 4.6 — costed side by side. The cheapest low-risk answer is not the obvious one.
Read the full report
Cheap small model > batching a frontier one
5M classification calls a month on GPT-5.4: batch it, or switch models?
For a high-volume, well-scoped task, the biggest lever is usually not the Batch API — it is the model. Batch, GPT-5.4 mini, Claude Haiku, Gemini Flash-Lite, and DeepSeek V4 Flash, all costed against the same workload.
Read the full report
Sonnet 4.6 if cost-driven; budget the rewrite
OpenAI → Anthropic for a coding assistant: worth the migration?
A $528k/yr coding-assistant bill on GPT-5.5, modeled against Claude Sonnet 4.6 and Opus 4.7 with prompt caching — plus the real engineering cost of the SDK and prompt-convention rewrite the switch requires.
Read the full report

Every figure is computed from the same pricing dataset that powers the model comparison calculators, verified 2026-05-27 against official provider pricing pages. Savings shown are list-price arithmetic on the stated workload — useful for planning, not a guaranteed invoice. Always validate model quality on your own evaluation set before switching.