So I am exploring option to translate my web app using AI. Usually I use Claude Code for that and result seems to be good enough. Some people meanwhile mentioned that DeepL is better in translating to many European languages. Actually research confirmed that.

What I have found that DeepL on free plan gives quite generous API access. I have created DeepL CLI with agentic skill and tried to use that for some time.

However my impression is that DeepL is not that good in translating software. I have even tried to give as much as possible context to it (DeepL does not charge for it), but it has failed to translated some parts correctly. Sometimes even single word was translated incorrectly. Even bigger problem is if we need to translate strings that has variables or counts. There is no native reliable way not to translate them - I don’t even speak about it being correct.

Final resolution:

• DeepL is good enough in translating longer texts (e.g. articles). I can reliably evaluate this only for English to Lithuanian translations, but that’s more than nothing. Since DeepL has generous free plan some money can be saved here.

• I am pretty sure that free plan of DeepL uses lower priority servers (service was not available once for me). I guess this means that DeepL on free plan is giving worse results. No idea…