Wonder vs GPT-6 Astra: Benchmark Comparison
Wonder has the higher score on 6 of 19 shared benchmarks; GPT-6 Astra leads on 13.
The largest observed score gap is 40.00 pts on Terminal-Bench-Science , where GPT-6 Astra leads.
Reported ±1 standard-error ranges overlap on 5 of 18 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 | Wonder | GPT-6 Astra | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 61.06% ±1.07 | 63.13% ±1.19 | 2.06 pts | Reported ±1 SE ranges overlap |
| Vals Science Index | 39.12% ±1.96 | 65.09% ±2.26 | 25.97 pts | Reported ±1 SE ranges do not overlap |
| Harvey's Legal Agent Benchmark | 17.08% ±3.10 | 5.42% ±1.17 | 11.67 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 50.48% ±3.48 | 39.42% ±3.40 | 11.06 pts | Reported ±1 SE ranges do not overlap |
| EMB | 67.08% ±2.89 | 71.70% ±2.48 | 4.62 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 56.64% ±0.38 | 53.54% ±2.08 | 3.10 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 36.27% ±3.47 | 20.72% ±2.92 | 15.55 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 84.00% ±3.69 | 99.00% ±1.00 | 15.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 67.04% ±0.37 | 79.26% ±0.98 | 12.22 pts | Reported ±1 SE ranges do not overlap |
| MysteryMechanism | 27.48% ±3.00 | 53.15% ±3.36 | 25.68 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench-Science | 22.86% ±5.05 | 62.86% ±5.82 | 40.00 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 62.55% ±4.35 | 67.74% ±4.22 | 5.19 pts | Reported ±1 SE ranges overlap |
| IOI | 66.94% ±4.91 | 100.00% ±0.00 | 33.06 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench Multimodal | 45.63% ±2.28 | 43.33% ±2.26 | 2.29 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 49.49% ±2.20 | 59.60% ±4.40 | 10.10 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 26.55% ±4.33 | 27.64% ±4.08 | 1.09 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 83.09% ±2.49 | 89.59% ±2.17 | 6.50 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 80.36% ±5.13 | 41.07% ±2.56 | 39.28 pts | Reported ±1 SE ranges do not overlap |
| CUA-bench | 7.17% | 19.17% | 12.00 pts | Uncertainty comparison unavailable |
Performance by category
| Category | Wonder average | GPT-6 Astra average |
|---|---|---|
| Legal | 33.78% | 22.42% |
| Finance | 53.33% | 48.65% |
| Math | 84.00% | 99.00% |
| Science | 39.12% | 65.09% |
| Coding | 55.71% | 64.65% |
| Cyber | 80.36% | 41.07% |
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 | Wonder cost | GPT-6 Astra cost | Wonder latency | GPT-6 Astra latency |
|---|---|---|---|---|
| Vals Index | $11.43 | $18.46 | 37m08s | 27m48s |
| Vals Science Index | $8.74 | $8.03 | 43m08s | 40m37s |
| Harvey's Legal Agent Benchmark | $7.08 | $26.16 | 22m50s | 24m49s |
| Legal Research Bench | $3.98 | $10.58 | 15m27s | 24m51s |
| EMB | $5.60 | $5.83 | 17m14s | 13m24s |
| Finance Agent (v2) | $2.67 | $6.82 | 6m09s | 11m33s |
| Tax Agent Bench | $1.88 | $5.80 | 8m13s | 15m01s |
| ProofBench v1.1 | $1.31 | $1.71 | 15m29s | 5m08s |
| BioMysteryBench | $3.23 | $1.72 | 18m02s | 4m46s |
| MysteryMechanism | $3.39 | $1.56 | 18m17s | 8m09s |
| Terminal-Bench-Science | $19.61 | $20.80 | 1h33m | 1h49m |
| Code Migration | $69.54 | $44.36 | 2h51m | 58m30s |
| IOI | $4.54 | $6.50 | 25m02s | 31m40s |
| SWE-bench Multimodal | $1.70 | $1.81 | 9m44s | 7m03s |
| Terminal-Bench 4.0 | $14.66 | $9.58 | 54m41s | 35m42s |
| Vibe Code Bench 1-100 | $13.72 | $84.72 | 31m08s | 3h26m |
| Vibe Code Bench v1.1 | $1.80 | $38.51 | 5m50s | 39m04s |
| CyberBench v1.1 | $3.66 | $1.84 | 13m45s | 4m10s |
| CUA-bench | $597.21 | $1821.37 | 30.56s | 32.09s |
Results available only for Wonder
- Public Benefits Bench v1.1
Results available only for GPT-6 Astra
- Vals RSI Index
- MedCode
- MedScribe
- Robo-bench
- SAGE
- ProgramBench
- SRE Bench
- Time Horizon Index: KSP