The short version: use GPT-5.5 where you already have it through ChatGPT or Codex. For API work, I would still default to GPT-5.4 for most business applications at the moment. DeepSeek belongs in the conversation too, but more as a low-cost option for tasks where review is already part of the workflow.
Use GPT-5.5 where the subscription already covers it
If you are working in ChatGPT or Codex and GPT-5.5 is included in your subscription, it is the model I would use first. It is stronger for coding, longer tasks, planning and the kind of back-and-forth work where the model has to keep context and recover from ambiguity.
That is especially true in Codex. When the cost is bundled into the subscription experience, the question is less about token pricing and more about whether you want the best available help for the task in front of you. Most of the time, you do.
For API work, GPT-5.4 is still the default I would choose
API projects are different because every call has a direct cost. GPT-5.5 may be the better model, but GPT-5.4 is around half the standard API price and is close enough for a lot of production work.
My current rule of thumb is to treat GPT-5.4 as roughly 90% as useful as GPT-5.5 for many coding, support, extraction, drafting and business workflow tasks. That is not a benchmark claim. It is a practical buying heuristic: when the last bit of quality is not worth double the model cost, start with GPT-5.4.
Start with GPT-5.4 in the API, then move specific high-value or failure-prone tasks to GPT-5.5 when the extra quality clearly pays for itself.
Where DeepSeek fits
DeepSeek is much cheaper again. For some workloads, the price difference is large enough that it deserves testing, especially for first drafts, classification, summarisation, internal tooling and jobs where a human or a stronger model will review the result anyway.
I would treat DeepSeek as roughly 70% as useful as GPT-5.4 for this kind of work, with the important caveat that it can make more mistakes and needs more checking. The trade-off can still make sense because DeepSeek pricing is less than a quarter of GPT-5.4's standard API price.
The decision rule
Use GPT-5.5 when the quality matters most, the subscription already covers it, or the task is complex enough that a stronger model will save real time. Use GPT-5.4 as the normal API default when you want strong results without paying top-tier API pricing for every request.
Use DeepSeek when cost matters more than perfect accuracy, when the task is low-risk, or when you already have a review step. Do not use the cheapest model for decisions where a subtle mistake would be expensive, embarrassing or hard to detect.
Check pricing before you lock it in
Model prices and availability change quickly. Before building costs into a client quote, software budget or automation plan, check the current official pricing pages and test the models on your own prompts.
Official references: OpenAI's GPT-5.5 launch note, OpenAI API pricing and DeepSeek API pricing.