31B > 235B? Gemma 4 31B vs Qwen3 235B on quality metrics

Posted by celsowm@reddit | LocalLLaMA | View on Reddit | 11 comments

Reply to Post

11 Comments

[-]

ttkciar@reddit

Violates Rule 3: Low-effort. If you are going to post about benchmarks, you need to show more than a screenshot, and provide analysis which genuinely improves the community's understanding.

[-]

DinoZavr@reddit

may i please wonder why not to compare with Qwen3.5 dense model 27B ?

[-]

celsowm@reddit (OP)

because we are currently using 235 for legal draft but we are considering to migrate to gemma4

[-]

ParaboloidalCrest@reddit

How dare you share findings relevant to your use-case instead of pleasing the lazyLlamas?! Seriously, thanks for sharing and may god gives you patience to withstand entitled requests coming your way.

[-]

StupidScaredSquirrel@reddit

Because it would be fair and uninteresting since we know qwen would win at everything except some creative writing obscure benchmark.

[-]

Adventurous-Paper566@reddit

Gemma gagnerait haut la main pour la compréhension des différentes langues européennes également.

[-]

StupidScaredSquirrel@reddit

Maybe for niche languages I'd agree. French, spanish ? I don't think so. But you're right that it would be interesting to see them neck to neck in different languages.

[-]

As a native French speaker, I've noticed that Qwen misses some of these subtleties. It struggles to grasp the colloquial meanings of words like 'écrémage' (weeding out) or 'boucherie' (a slaughter/total mess), for instance.

[-]

DinoZavr@reddit

i am still testing new Gemma. as it's predecessor new Gemma is absolutely fantastic with image captioning and writing. For language translation and coding Qwen3.5 27B is much better than anything else of compartible size. Though i still stated testing and i am, probably, biased towards Qwen, as i am still so much impressed by Qwen3.5-27B's capabilities. Too early to judge, though i was waiting Gemma4 to emerge in April..May and this eventually has happened.

[-]

BumblebeeParty6389@reddit

Gemma is a new model and they need to show it's not useless