DeepSeek V4.1 Flash vs Mercury 3
Too close to call. On the work we tested, DeepSeek V4.1 Flash and Mercury 3 can't be told apart. Mercury 3 is the cheaper one, slightly.
GrindBench-PubSec
BDeepSeek V4.1 FlashDeepSeek64% 59–70%0.43¢ a task · 300 tasksBMercury 3Inception61% 56–67%0.53¢ a task · 300 tasks
Too close to call here: their ranges overlap, so on another set of tasks the order could flip.
| Kind of work | DeepSeek V4.1 Flash | Mercury 3 |
|---|---|---|
| ATIP redaction | 20% | 11% |
| Procurement evaluation | 78% | 86% |
| Performance analysis | 63% | 54% |
| Bilingual equivalence | 95% | 97% |
GrindBench-Banking
BDeepSeek V4.1 FlashDeepSeek62% 48–73%1.6¢ a task · 60 tasksCMercury 3Inception52% 38–63%1.4¢ 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 | DeepSeek V4.1 Flash | Mercury 3 |
|---|---|---|
| Lending & mortgages | 60% | 50% |
| AML & KYC | 70% | 40% |
| Deposits & payments | 40% | 40% |
| Complaints & conduct | 70% | 70% |
| Credit risk | 80% | 60% |
| Capital & liquidity | 50% | 50% |
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.