Ember-1 vs Step 5 Preview: Benchmark Comparison
Ember-1 has the higher score on 9 of 11 shared benchmarks; Step 5 Preview leads on 2.
The largest observed score gap is 19.00 pts on ProofBench v1.1 , where Ember-1 leads.
Reported ±1 standard-error ranges overlap on 7 of 11 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 | Ember-1 | Step 5 Preview | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Science Index | 31.76% ±1.66 | 29.58% ±1.10 | 2.18 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 44.71% ±3.46 | 40.38% ±3.41 | 4.33 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 31.08% ±3.34 | 27.27% ±3.11 | 3.80 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 61.00% ±4.90 | 42.00% ±4.96 | 19.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 67.04% ±3.03 | 67.41% ±0.37 | 0.37 pts | Reported ±1 SE ranges overlap |
| MysteryMechanism | 22.52% ±2.81 | 18.47% ±2.61 | 4.05 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 5.71% ±2.79 | 2.86% ±2.01 | 2.86 pts | Reported ±1 SE ranges overlap |
| IOI | 46.33% ±9.16 | 38.06% ±6.69 | 8.28 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 20.71% ±3.64 | 31.82% ±2.62 | 11.11 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 83.87% ±3.15 | 70.40% ±4.27 | 13.47 pts | Reported ±1 SE ranges do not overlap |
| Public Benefits Bench v1.1 | 66.17% ±1.23 | 58.12% ±1.28 | 8.05 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Ember-1 average | Step 5 Preview average |
|---|---|---|
| Legal | 44.71% | 40.38% |
| Finance | 31.08% | 27.27% |
| Math | 61.00% | 42.00% |
| Science | 31.76% | 29.58% |
| Coding | 50.30% | 46.76% |
| Social Mobility | 66.17% | 58.12% |
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 | Ember-1 cost | Step 5 Preview cost | Ember-1 latency | Step 5 Preview latency |
|---|---|---|---|---|
| Vals Science Index | $10.99 | $1.78 | 1h18m | 1h07m |
| Legal Research Bench | $3.67 | $0.84 | 38m34s | 36m08s |
| Tax Agent Bench | $2.97 | $0.55 | 30m46s | 21m29s |
| ProofBench v1.1 | $1.74 | $0.71 | 21m09s | 38m34s |
| BioMysteryBench | $0.91 | $0.49 | 9m26s | 21m35s |
| MysteryMechanism | $1.29 | $0.60 | 20m15s | 29m20s |
| Terminal-Bench-Science | $30.76 | $4.25 | 3h23m | 2h30m |
| IOI | $24.46 | $2.03 | 1h47m | 1h20m |
| Terminal-Bench 4.0 | $19.48 | $3.11 | 1h55m | 1h58m |
| Vibe Code Bench v1.1 | $8.56 | $5.86 | 39m18s | 1h31m |
| Public Benefits Bench v1.1 | $1.02 | $0.50 | 13m21s | 27m19s |
Results available only for Ember-1
- CyberBench v1.1
Results available only for Step 5 Preview
- Vals Finance Index
- EMB v1.1
- Finance Agent v2.1
- Code Migration v1.1
- SWE-bench Multimodal
- SRE Bench