Every comparison of these three assumes the same student: someone reading, thinking, and writing in one language. That’s not most of the world. Across the Middle East, and in plenty of other places, students sit through lectures in English, take notes half in English and half in their own language, and then try to study from both at once. None of the popular “ChatGPT vs Claude vs Gemini for students” posts — including the two already on this site — test for that. This one does.

I ran a semester’s worth of real coursework through all three, deliberately mixing languages the way I actually study: English textbook, Arabic explanation, English exam. Here’s where each one helped and where each one quietly made things worse.

Why this is a different test than “which one is smartest”

A model can be excellent at English and excellent at Arabic separately and still be bad at the thing students actually need: moving between the two without losing precision. Two failure modes show up constantly in mixed-language studying:

Neither of these shows up if you only test in one language, which is exactly why the generic comparisons miss it.

Test 1: Explain it in Arabic, keep the English term intact

I fed each model a page from an English-language networking textbook (subnetting) and asked, in Arabic, for the explanation in Arabic — but told it to keep technical terms in English, which is how these subjects are actually taught and tested here.

ChatGPT handled the instruction well on the first try and kept doing it consistently across a long back-and-forth — it didn’t drift back into full-English explanations even after ten or so follow-up questions. Gemini did the same thing but occasionally translated a term I’d explicitly asked it to leave in English, which is a small mistake with a real cost if you copy it straight into exam prep. Claude gave the most precise Arabic explanation of the three — clearer sentence structure, less awkward phrasing — but was the most likely to slip and answer a follow-up question in English after a few turns, especially once the conversation got technical.

Test 2: Notes in two languages, one summary

Real student notes are messy: an English slide screenshot, a voice-note transcript in Arabic, a WhatsApp message from a classmate. I pasted a mix of English and Arabic notes on the same topic and asked for one merged summary.

This is where the differences got bigger than in Test 1. ChatGPT merged the two sources but leaned English by default unless told otherwise — fine if that’s your exam language, annoying if you wanted to review in Arabic. Gemini was the most willing to keep the summary bilingual, labeling which part came from which source, which actually matched how I wanted to review. Claude produced the best-organized summary of the three but needed to be told explicitly which output language to use — left to its own judgment, it guessed based on whichever language had more words in the input, and guessed wrong about a third of the time in my test.

Test 3: Practice questions in the exam’s actual language

Understanding a concept in Arabic doesn’t help on exam day if you’ve never seen it phrased the way the exam phrases it. I asked all three to generate practice questions in the same register and phrasing style as past exam questions, after being shown two or three real examples.

ChatGPT was the most reliable at matching the exact phrasing style once given examples — closest to what a professor would actually write. Claude’s practice questions were well-written but tended to be more elaborate than the source exams, which is nice for understanding but not great for practicing under real time pressure. Gemini’s questions were serviceable but the least consistent match to the tone of the sample questions.

What actually matters if this is your situation

ChatGPTClaudeGemini
Sticks to the language you asked in, even in long chatsMost consistentCan drift to English after several turnsConsistent
Keeps specified technical terms untranslatedReliableReliableOccasionally translates them anyway
Merging mixed-language notesDefaults to English unless toldBest organized, but guesses the output languageBest at labeling and keeping both languages
Matching real exam phrasing for practice questionsBest matchMore elaborate than the sourceLeast consistent match

None of the three is built specifically for bilingual studying — this is a side effect of general capability, not a designed feature, and it shows. All three will drift, mistranslate, or default to the wrong language at some point in a long study session. The fix that worked consistently across all three: state your target language explicitly in every prompt rather than trusting the model to infer it from your notes, and double-check any technical term it translates against your actual course glossary before it goes into a flashcard.

VERDICT

If you're studying material in one language and thinking in another, don't pick a model on general reputation — pick based on which failure mode costs you less. Use ChatGPT if you mainly need consistent language behavior over a long study session and exam-realistic practice questions. Use Gemini if you're merging notes that are already a real mix of two languages and want to keep both visible rather than collapsed into one. Use Claude if precision and organization of the explanation itself matters more to you than the model reliably staying in the language you started in — just watch it and redirect it back when it drifts. Whichever you use, never trust a translated technical term without checking it against your own course material first.