Hy4 Preview vs GLM 5.3: Benchmark Comparison
Hy4 Preview has the higher score on 6 of 18 shared benchmarks; GLM 5.3 leads on 12.
The largest observed score gap is 30.81 pts on Terminal-Bench 4.0 , where GLM 5.3 leads.
Reported ±1 standard-error ranges overlap on 10 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 | Hy4 Preview | GLM 5.3 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 49.94% ±1.15 | 53.51% ±1.30 | 3.57 pts | Reported ±1 SE ranges do not overlap |
| Harvey's Legal Agent Benchmark | 9.17% ±2.00 | 8.33% ±2.00 | 0.83 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 45.19% ±3.46 | 49.04% ±3.48 | 3.85 pts | Reported ±1 SE ranges overlap |
| LegalBench | 83.76% ±0.41 | 84.84% ±0.40 | 1.08 pts | Reported ±1 SE ranges do not overlap |
| EMB | 58.08% ±3.14 | 56.34% ±3.32 | 1.73 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 55.06% ±0.31 | 55.84% ±2.07 | 0.78 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 63.71% ±3.26 | 73.09% ±2.97 | 9.38 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 43.25% ±2.13 | 42.86% ±2.11 | 0.38 pts | Reported ±1 SE ranges overlap |
| MedScribe | 83.60% ±2.06 | 88.81% ±2.00 | 5.21 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 75.00% ±4.35 | 49.00% ±5.02 | 26.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench Science | 1.43% ±1.43 | 5.71% ±2.79 | 4.29 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 47.43% ±4.27 | 44.22% ±4.29 | 3.21 pts | Reported ±1 SE ranges overlap |
| IOI | 59.33% ±4.60 | 68.44% ±7.48 | 9.11 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 1.50% ±0.86 | 1.50 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 8.08% ±1.34 | 38.89% ±1.82 | 30.81 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 77.48% ±4.04 | 78.13% ±3.94 | 0.65 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 65.36% ±5.55 | 72.08% ±5.41 | 6.73 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 68.61% ±1.21 | 68.54% ±1.21 | 0.07 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | Hy4 Preview average | GLM 5.3 average |
|---|---|---|
| Legal | 46.04% | 47.40% |
| Finance | 58.95% | 61.76% |
| Healthcare | 63.42% | 65.84% |
| Math | 75.00% | 49.00% |
| Science | 1.43% | 5.71% |
| Coding | 38.47% | 46.24% |
| Cyber | 65.36% | 72.08% |
| Social Mobility | 68.61% | 68.54% |
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 | Hy4 Preview cost | GLM 5.3 cost | Hy4 Preview latency | GLM 5.3 latency |
|---|---|---|---|---|
| Vals Index | $1.44 | $7.25 | 1h01m | 1h13m |
| Harvey's Legal Agent Benchmark | $0.98 | $4.30 | 29m25s | 42m38s |
| Legal Research Bench | $0.85 | $2.24 | 1h04m | 51m50s |
| LegalBench | N/A | N/A | 90.56s | 22.48s |
| EMB | $0.79 | $3.79 | 30m39s | 36m12s |
| Finance Agent (v2) | $0.59 | $1.07 | 23m45s | 15m50s |
| Tax Agent Bench | $0.59 | $1.80 | 41m04s | 50m53s |
| MedCode | N/A | N/A | 6m39s | 2m49s |
| MedScribe | N/A | N/A | 5m50s | 2m02s |
| ProofBench v1.1 | $0.34 | $2.08 | 22m54s | 42m23s |
| Terminal-Bench Science | $2.14 | $15.63 | 3h48m | 2h45m |
| Code Migration | $3.41 | $24.91 | 2h13m | 3h47m |
| IOI | $1.35 | $7.67 | 1h01m | 1h46m |
| ProgramBench | $12.81 | $21.96 | 2h10m | 4h16m |
| Terminal-Bench 4.0 | $2.10 | $9.37 | 2h10m | 1h23m |
| Vibe Code Bench v1.1 | $2.18 | $12.45 | 41m33s | 1h04m |
| CyberBench v1.1 | $0.49 | $2.69 | 36m43s | 30m18s |
| Public Benefits Bench v1.1 | $0.25 | $0.95 | 1h03m | 35m51s |
Results available only for Hy4 Preview
- BioMysteryBench
- SRE Bench
Results available only for GLM 5.3
- Vals RSI Index
- TaxEval v2
- MysteryMechanism
- GPQA Diamond
- MMLU Pro
- LiveCodeBench
- SkillsBench
- SWE-bench
- Vibe Code Bench 1-100