DeepSeek V4 Flash (Non-reasoning): benchmark results
DeepSeek V4 Flash non-reasoning evaluation mode. Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1635 ± 10, rank #582 of 2168 rated models, from 124 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Cloud-OpsBench (OnlineBoutique) | 76 | Joint RCA accuracy (%): share of episodes whose predicted fa | 100 |
| Cloud-OpsBench (OnlineBoutique) - Component Accuracy | 88 | Component accuracy (%): share of episodes whose predicted fa | 100 |
| Cloud-OpsBench (OnlineBoutique) - Fault-Type Accuracy | 77 | Fault-type accuracy (%): share of episodes whose predicted f | 100 |
| JANUS - Emphasis | -0.03 | Emphasis distortion, signed from -2 to 2: change in the atte | 100 |
| UGI Leaderboard | 59.21 | UGI Score | 98.6 |
| Tau2-Bench Telecom | 94.44 | Success Rate (%) | 92.9 |
| JANUS - Framing | 0.17 | Framing distortion, signed from -2 to 2: change in the mean | 90.9 |
| UGI - Writing | 54.58 | Writing Score | 89.2 |
| Cloud-OpsBench (TrainTicket) - Component Accuracy | 69 | Component accuracy (%): share of episodes whose predicted fa | 88.9 |
| Cloud-OpsBench (TrainTicket) - Evidence Closure Rate | 24 | Evidence closure rate (%): share of episodes whose trajector | 88.9 |
| Cloud-OpsBench (TrainTicket) - Evidence-Order Consistency | 56 | Evidence-order consistency (%): mean share of milestone grou | 88.9 |
| Cloud-OpsBench (TrainTicket) - Milestone Coverage | 62 | Milestone coverage (%): mean share of the case's expert diag | 88.9 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.