How this calculator works
The LLM Vendor Migration Cost Calculator estimates the true one-time cost of moving your application from one large language model provider to another, rather than just comparing per-token API prices. It adds up three main drivers: re-embedding your document corpus (your number of chunks multiplied by average tokens per chunk, priced at the new provider's embedding rate per million tokens), re-running fine-tuning (training tokens priced at the new fine-tune rate per million), and engineering integration time (billed hours to rewire SDKs, prompts, and tests). Each embedding and fine-tuning component can be toggled off if that step isn't needed for your migration.
The key insight is that headline API savings often understate what switching actually costs. Re-embedding a large knowledge base and paying engineers to re-integrate can dwarf a small per-token discount, and those costs are paid up front before any savings accrue. Use this tool to estimate a realistic break-even point: divide the total switching cost by your expected monthly savings to see how many months the new provider must run before the move pays for itself. If that horizon is long or your volume is uncertain, the cheaper sticker price may not justify the migration.
Frequently asked questions
What are the real costs of switching LLM providers?
Beyond API price differences, switching costs include: re-embedding your entire knowledge base (embeddings are not compatible across providers), re-running any fine-tuning jobs, engineering time to rewrite prompts and API calls, integration testing, and regression testing. These one-time costs can easily exceed 3–6 months of API savings.
Are embeddings portable between providers?
No. Vector embeddings from OpenAI, Anthropic, Cohere, and others live in different vector spaces and are not interchangeable. Switching embedding providers means re-embedding every document in your knowledge base and rebuilding your vector index. This is often the largest hidden migration cost.
How long does it take to recoup LLM migration costs?
Payback period = total migration cost ÷ monthly savings from the new provider. If migration costs $15,000 in engineering time and the new model saves $800/month, payback is about 19 months. This calculator shows that math so you can decide if the switch is worth it.
Do prompts need to be rewritten when changing LLM providers?
Often yes. Prompt sensitivity varies significantly between models. A prompt tuned for GPT-4 may need substantial reworking for Claude, and vice versa. Budget at least 20–30% of integration time for prompt engineering and regression testing.
What is the cheapest way to switch LLM providers?
Use an AI gateway or router (LiteLLM, Portkey, Helicone) that abstracts the provider API. This reduces integration engineering to changing a model string, but you still need to re-embed any RAG knowledge base and re-run fine-tuning jobs.