Head to head: Google: Gemini 3.6 Flash vs GLM 5.2

Google's Gemini 3.6 Flash defeated GLM 5.2 by a score of 68.0 to 53.0 in a head-to-head comparison of instruction-following ability, winning 6 of 8 tasks with one tie and 96% confidence. The test, conducted by running 8 fresh text tasks judged twice by GPT-5.4 to cancel position bias, showed Gemini was more reliable at handling constraint-heavy prompts while GLM often violated exact formatting or letter-count rules.

Gemini 3.6 Flash wins this head-to-head cleanly: 68.0 to 53.0 overall, with 96% confidence . That’s not a marginal edge or a judge-quirk result. It’s a real separation, backed up by a 6–1 task record with one tie . The pattern is straightforward: Gemini was the more dependable model when prompts got fussy, formal, or adversarial. It handled constraint-heavy asks better across the board — from the lowercase-and-letter-count resume, to multiple limerick tasks, to the tightly formatted math explanation. In several cases, GLM 5.2 had decent underlying content, but lost because it missed the actual assignment: violating letter-count rules, weakening required poetic form, or tripping over exact formatting constraints. GLM’s lone win is telling. It came on a haiku-style instruction-following task where it better satisfied the full bundle of requirements — title formatting, lowercase, Ireland imagery, and especially the required density of the letter g. So this is not a case of total nonperformance; GLM can absolutely land a constrained creative prompt when the pieces line up. But across the full set, that was the exception, not the rule. What really hurts GLM here is that its losses were often avoidable. Judges repeatedly noted that it was sometimes more detailed or conceptually solid, yet still came up short because it failed exact constraints or produced clumsier execution. Gemini, by contrast, was consistently the model that both understood the brief and actually delivered it in the requested form. Final call: Gemini 3.6 Flash is the clear winner. If you care about reliable instruction following rather than occasional flashes of quality wrapped in preventable misses, this matchup isn’t close. How they were tested We ran 8 fresh text tasks, generated on the fly for this matchup so neither model could prepare in advance, and had gpt-5.4 score each one. To cancel position bias, every task was judged twice — once in each presentation order — and every number reported here, including the headline totals, is the average of both passes. Google: Gemini 3.6 Flash scored 68.0 to GLM 5.2's 53.0. 1. IFEval · instruction following can you write a resume for helene? Answer with lowercase letters. Make sure the letter n appears less than 7 times. Winner: Google: Gemini 3.6 Flash — Model A better satisfies the key constraint: it is entirely in lowercase and appears to keep the letter 'n' under 7 total uses, while still producing a plausible resume for helene. Model B is more detailed, but it clearly violates the letter-count constraint many times, so its stronger content does not outweigh poorer instruction following. Second judge pass, order swapped — scores are the average of both: Model A better satisfies the key constraint about keeping the letter n under 7 occurrences while still providing a plausible resume in lowercase. Model B is more complete as a resume, but it clearly violates the letter-count instruction by using many instances of the letter n. 2. IFEval · instruction following write a haiku about ireland in the style of basho. include a title wrapped in double angular brackets, i.e. <<title . your entire response should be in lowercase english and contain the letter g at least 8 times. Winner: GLM 5.2 — Model B better satisfies the constraints: it has the required double-angle-bracket title, stays lowercase, clearly evokes ireland, and includes at least eight instances of the letter g. Model A is evocative but falls short on the letter-count requirement and is slightly less polished as a basho-like haiku. Second judge pass, order swapped — scores are the average of both: Model B better satisfies the prompt overall: it clearly evokes ireland, stays fully lowercase, uses the required title format, and includes well over eight instances of the letter g while delivering a stronger basho-like natural stillness. Model A is also strong, but it feels a bit more literal and less refined poetically, and its ireland reference is more indirect. 3. IFEval · instruction following Write a limerick about a guy from Nantucket, use notations to express it, and use at least 2 words with all capital letters. Winner: Google: Gemini 3.6 Flash — Model A clearly delivers a proper limerick about a guy from Nantucket, uses explicit notation devices throughout, and includes multiple all-caps words. Model B includes all-caps words and a Nantucket setup, but its rhyme/meter are weaker, the ending is awkward, and the use of notation is more loosely explained than actually integrated. Second judge pass, order swapped — scores are the average of both: Model A better satisfies the prompt by delivering a clear limerick about a guy from Nantucket while explicitly using multiple forms of notation and at least two all-caps words. Model B includes all-caps words, but it is less clearly a proper limerick and leans on explanatory prose outside the poem rather than integrating notation into the verse itself. 4. IFEval · instruction following Create a product description for a product that will help me to stop snoring. Use all lowercase letters. Winner: Tie — Model B better satisfies the prompt with a concise, polished product description entirely in lowercase and with strong marketing clarity. Model A also follows the lowercase instruction, but it is slightly wordier and makes a more absolute claim about "instantly eliminating" snoring, which is less careful and lowers correctness. Second judge pass, order swapped — scores are the average of both: Model A better fulfills the prompt with a polished, persuasive product description that stays entirely in lowercase and presents the product benefits more clearly and professionally. Model B also follows the lowercase instruction, but Model A is slightly stronger in specificity, structure, and overall marketing quality. 5. IFEval · instruction following Which of the following is a better way to describe supporting ecological landscapes: A the practice of using the natural features of a landscape to create a more sustainable environment, or B the practice of using the natural features of a landscape to create a more aesthetically pleasing environment? Your response should be in English, all capital letters, contain no commas, and be fewer than 16 sentences. Winner: Google: Gemini 3.6 Flash — Model A is better because it fully follows the formatting constraints while giving a clear and correct explanation that option A is more accurate. Model B is also correct and well written but it appears as a single run-on sentence which hurts readability and writing quality. Second judge pass, order swapped — scores are the average of both: Model A is better because it is correct and clearly argued while staying within the formatting constraints. Model B is also correct and follows the capitalization and comma rules but its single extremely long sentence hurts readability and writing quality. 6. IFEval · instruction following Write a professional haiku in English for moms about an inspirational chorus teacher. It should include the phrase "singing is life" and be in all lowercase letters. No capital letters are allowed. Winner: Google: Gemini 3.6 Flash — Model A better fits the prompt by clearly centering an inspirational chorus teacher and moms while including the exact phrase "singing is life" in all lowercase. Model B also follows the constraints, but it is a bit less specific about the chorus-teacher role and feels slightly more generic. Second judge pass, order swapped — scores are the average of both: Model A better fits the prompt by clearly centering an inspirational chorus teacher and moms while including the exact phrase "singing is life" in all lowercase. Model B is strong, but it is less explicitly about a chorus teacher and feels slightly less specific to the requested theme. 7. IFEval · instruction following Write a limerick about a guy named Dave that is funny to moms. The limerick should end with the phrase "Yes Mom, I am Dave." Do not say anything after the limerick. Winner: Google: Gemini 3.6 Flash — Model A cleanly delivers a five-line limerick about Dave, ends exactly with the required phrase, and is mildly mom-friendly and coherent. Model B does end with the required phrase, but it breaks the limerick form more noticeably, has awkward wording like "resembled an eel," and reads less coherently and less humorously for the requested audience. Second judge pass, order swapped — scores are the average of both: Model A is a proper five-line limerick that cleanly ends with the exact required phrase and has a clearer, more coherent joke. Model B fails the limerick form by effectively splitting the ending into extra lines and is less polished and less funny. 8. IFEval · instruction following What's different between "the set of all sets that are not members of themselves" and "the set of all sets"? Use mathematical notations in your answer. Be chatty while explaining. There should be exactly 2 paragraphs each separated by two new lines in your response. Paragraph 1 must start with the word booster. Winner: Google: Gemini 3.6 Flash — Model A better satisfies the prompt while remaining mathematically strong: it has exactly two paragraphs, uses notation well, and gives a more nuanced comparison, including the point that under Foundation/Regularity one gets \forall x, x\notin x , so the two descriptions would coincide extensionally in ZFC even though neither is a set. Model B is also good, but it is slightly less precise because it asserts the universal set would include itself and frames that as part of the definition rather than a consequence of assuming it is a set containing all sets, and its explanation is a bit less subtle overall. Second judge pass, order swapped — scores are the average of both: Model A better satisfies the formatting and tone requirements: it has exactly two paragraphs, uses mathematical notation, and starts paragraph 1 with the required word in lowercase. Model B is mathematically solid overall, but it violates the instruction because it begins with "Booster" rather than the exact word "booster"; Model A has a minor overstatement in saying the collections are "classified as proper classes" in ZFC, but it is still the stronger response overall. See every prompt and the full side-by-side outputs in the interactive Head-to-Head /head-to-head/head-to-head-google-gemini-3-6-flash-vs-glm-5-2 .