DeepSeek V4.1 Flash vs Ember-1: Benchmark Comparison
DeepSeek V4.1 Flash has the higher score on 3 of 12 shared benchmarks; Ember-1 leads on 9.
The largest observed score gap is 7.00 pts on ProofBench v1.1 , where Ember-1 leads.
Reported ±1 standard-error ranges overlap on 11 of 12 shared benchmarks with comparable uncertainty data. This is not a pairwise statistical significance test.
Shared benchmark results
Scores and reported standard errors come from the latest visible Vals benchmark version available for both models.
| Benchmark | DeepSeek V4.1 Flash | Ember-1 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Science Index | 29.65% ±0.99 | 31.76% ±1.66 | 2.11 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 41.35% ±3.42 | 44.71% ±3.46 | 3.37 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 26.63% ±3.14 | 31.08% ±3.34 | 4.45 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 54.00% ±5.01 | 61.00% ±4.90 | 7.00 pts | Reported ±1 SE ranges overlap |
| BioMysteryBench | 67.78% ±1.11 | 67.04% ±3.03 | 0.74 pts | Reported ±1 SE ranges overlap |
| MysteryMechanism | 21.17% ±2.75 | 22.52% ±2.81 | 1.35 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 0.00% ±0.00 | 5.71% ±2.79 | 5.71 pts | Reported ±1 SE ranges do not overlap |
| IOI | 40.28% ±2.56 | 46.33% ±9.16 | 6.05 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 19.70% ±1.75 | 20.71% ±3.64 | 1.01 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 84.74% ±2.89 | 83.87% ±3.15 | 0.87 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 73.69% ±5.48 | 69.52% ±5.56 | 4.17 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 64.28% ±1.25 | 66.17% ±1.23 | 1.90 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | DeepSeek V4.1 Flash average | Ember-1 average |
|---|---|---|
| Legal | 41.35% | 44.71% |
| Finance | 26.63% | 31.08% |
| Math | 54.00% | 61.00% |
| Science | 29.65% | 31.76% |
| Coding | 48.24% | 50.30% |
| Cyber | 73.69% | 69.52% |
| Social Mobility | 64.28% | 66.17% |
Cost and latency
Cost per test appears only where the benchmark reports it for both models. Latency is the measured completion time for that benchmark.
| Benchmark | DeepSeek V4.1 Flash cost | Ember-1 cost | DeepSeek V4.1 Flash latency | Ember-1 latency |
|---|---|---|---|---|
| Vals Science Index | $0.29 | $10.99 | 1h07m | 1h18m |
| Legal Research Bench | $0.25 | $3.67 | 11m42s | 38m34s |
| Tax Agent Bench | $0.13 | $2.97 | 5m49s | 30m46s |
| ProofBench v1.1 | $0.13 | $1.74 | 9m07s | 21m09s |
| BioMysteryBench | $0.23 | $0.91 | 14m50s | 9m26s |
| MysteryMechanism | $0.16 | $1.29 | 8m04s | 20m15s |
| Terminal-Bench-Science | $0.47 | $30.76 | 2h57m | 3h23m |
| IOI | $0.24 | $24.46 | 18m30s | 1h47m |
| Terminal-Bench 4.0 | $0.50 | $19.48 | 48m22s | 1h55m |
| Vibe Code Bench v1.1 | $0.41 | $8.56 | 15m28s | 39m18s |
| CyberBench v1.1 | $0.07 | $2.43 | 13m50s | 22m02s |
| Public Benefits Bench v1.1 | $0.07 | $1.02 | 9m59s | 13m21s |
Results available only for DeepSeek V4.1 Flash
- Vals Finance Index
- Vals Inference Provider Index
- LegalBench
- EMB v1.1
- Finance Agent v2.1
- MedCode
- MedScribe
- SAGE
- Code Migration v1.1
- ProgramBench
- SkillsBench
- SWE-bench Multimodal
- Vibe Code Bench 1-100
- SRE Bench
Results available only for Ember-1
None.