Gemma 4, Qwen3.5 split benchmark wins

The useful signal here is that open-weight frontiers are converging fast, so benchmark tables are increasingly about tradeoffs, not crowning a single winner.

• –The comparison is only as fair as the source data: both families publish their own benchmark numbers, so treat the table as directional rather than definitive.
• –Gemma 4 is positioned hard around local and edge deployment, so its value depends on whether you care more about on-device efficiency than absolute scale.
• –Qwen3.5 still looks like the broader platform play, with huge context, multilingual reach, and a mature tool-use story.
• –For builders, the decision point is likely to be latency, memory budget, languages, and agent reliability, not a single benchmark aggregate.