Ember-1 vs GPT-6 Luna: Benchmark Comparison
Ember-1 has the higher score on 9 of 12 shared benchmarks; GPT-6 Luna leads on 3.
The largest observed score gap is 14.42 pts on Legal Research Bench , where Ember-1 leads.
Reported ±1 standard-error ranges overlap on 7 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 | Ember-1 | GPT-6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Science Index | 31.76% ±1.66 | 28.38% ±1.48 | 3.38 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 44.71% ±3.46 | 30.29% ±3.19 | 14.42 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 31.08% ±3.34 | 19.95% ±2.87 | 11.13 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 61.00% ±4.90 | 64.00% ±4.82 | 3.00 pts | Reported ±1 SE ranges overlap |
| BioMysteryBench | 67.04% ±3.03 | 61.48% ±2.59 | 5.56 pts | Reported ±1 SE ranges overlap |
| MysteryMechanism | 22.52% ±2.81 | 19.37% ±2.66 | 3.15 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 5.71% ±2.79 | 4.29% ±2.44 | 1.43 pts | Reported ±1 SE ranges overlap |
| IOI | 46.33% ±9.16 | 55.56% ±8.87 | 9.22 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 20.71% ±3.64 | 13.64% ±1.51 | 7.07 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 83.87% ±3.15 | 81.65% ±3.38 | 2.22 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 69.52% ±5.56 | 76.25% ±5.29 | 6.73 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 66.17% ±1.23 | 57.65% ±1.28 | 8.53 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Ember-1 average | GPT-6 Luna average |
|---|---|---|
| Legal | 44.71% | 30.29% |
| Finance | 31.08% | 19.95% |
| Math | 61.00% | 64.00% |
| Science | 31.76% | 28.38% |
| Coding | 50.30% | 50.28% |
| Cyber | 69.52% | 76.25% |
| Social Mobility | 66.17% | 57.65% |
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 | GPT-6 Luna cost | Ember-1 latency | GPT-6 Luna latency |
|---|---|---|---|---|
| Vals Science Index | $10.99 | $0.11 | 1h18m | 30m15s |
| Legal Research Bench | $3.67 | $0.44 | 38m34s | 42m28s |
| Tax Agent Bench | $2.97 | $0.19 | 30m46s | 33m16s |
| ProofBench v1.1 | $1.74 | $0.04 | 21m09s | 7m31s |
| BioMysteryBench | $0.91 | $0.05 | 9m26s | 8m44s |
| MysteryMechanism | $1.29 | $0.03 | 20m15s | 6m01s |
| Terminal-Bench-Science | $30.76 | $0.24 | 3h23m | 1h16m |
| IOI | $24.46 | $0.20 | 1h47m | 35m47s |
| Terminal-Bench 4.0 | $19.48 | $0.35 | 1h55m | 32m46s |
| Vibe Code Bench v1.1 | $8.56 | $1.35 | 39m18s | 37m00s |
| CyberBench v1.1 | $2.43 | $0.13 | 22m02s | 15m50s |
| Public Benefits Bench v1.1 | $1.02 | $0.22 | 13m21s | 52m59s |
Results available only for Ember-1
None.
Results available only for GPT-6 Luna
- Vals Finance Index
- Vals Index
- Vals Legal Index
- HLAB-Vals-1.1
- ClaimBench
- EMB v1.1
- Finance Agent v2.1
- MedCode
- MedScribe
- SAGE
- Code Migration v1.1
- ProgramBench
- SWE-bench Multimodal
- SRE Bench