Native vision model using a causal encoder-decoder MoE architecture, with 8B active parameters for input and 16B for output. Replaces V4 Flash API aliases; V4 Pro service continues under updated September 14 guidance. Listed prices are peak rates; off-peak rates are half.
SI v3 uses weighted percentages from one Artificial Analysis configuration. Ranking requires all five benchmarks. Partial scores average only available weights. Scores are not comparable to SI v2 or a probability of reaching singularity.
Compared with 59 models released within six months of it, across 3 benchmarks where scores spread out.
| Benchmark | Score | Rank |
|---|---|---|
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 39.2% | #42 / 95 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisDeepSeek V4.1 Flash (Max) | 39.2% |
MMMU-Pro Artificial AnalysisDeepSeek V4.1 Flash (Max) | 77% |
Terminal-Bench 4.0 Artificial AnalysisDeepSeek V4.1 Flash (Max) | 26.8% |
SciCode Artificial AnalysisDeepSeek V4.1 Flash (Max) | 51.9% |
AA-LCR Artificial AnalysisDeepSeek V4.1 Flash (Max) | 84% |
CritPt Artificial AnalysisDeepSeek V4.1 Flash (Max) | 14.3% |
ITBench SRE Artificial AnalysisDeepSeek V4.1 Flash (Max) | 46.9% |