Qwen3.6 35B A3B vs Gemma 4 31B
Qwen3.6 35B A3B is ahead on some work, and they tie on the rest. Gemma 4 31B is the cheaper one, by 10×.
GrindBench-PubSec
DQwen3.6 35B A3BAlibaba (Qwen)33% 27–38%0.53¢ a task · 300 tasksDGemma 4 31BGoogle21% 16–26%0.036¢ a task · 300 tasks
Qwen3.6 35B A3B is ahead here by 12 points, more than chance explains.
| Kind of work | Qwen3.6 35B A3B | Gemma 4 31B |
|---|---|---|
| ATIP redaction | 0% | 0% |
| Procurement evaluation | 58% | 14% |
| Performance analysis | 11% | 1% |
| Bilingual equivalence | 85% | 90% |
GrindBench-Banking
FQwen3.6 35B A3BAlibaba (Qwen)3% 0–8%1.8¢ a task · 60 tasksFGemma 4 31BGoogle0% 0–0%0.19¢ a task · 60 tasks
Too close to call here: their ranges overlap, so on another set of tasks the order could flip.
| Kind of work | Qwen3.6 35B A3B | Gemma 4 31B |
|---|---|---|
| Lending & mortgages | 20% | 0% |
| AML & KYC | 0% | 0% |
| Deposits & payments | 0% | 0% |
| Complaints & conduct | 0% | 0% |
| Credit risk | 0% | 0% |
| Capital & liquidity | 0% | 0% |
Each kind of work here rests on 10 tasks per model, so a single row can easily be 20 points off. A lead of 10 points or more is in bold; read it as a hint, not a result.