Gemini 3.1 Pro Preview (02/26) vs GPT 5.4 Nano: Benchmark Comparison
Gemini 3.1 Pro Preview (02/26) has the higher score on 16 of 18 shared benchmarks; GPT 5.4 Nano leads on 1.
The largest observed score gap is 18.03 pts on MedCode , where Gemini 3.1 Pro Preview (02/26) leads.
Reported ±1 standard-error ranges overlap on 4 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 | Gemini 3.1 Pro Preview (02/26) | GPT 5.4 Nano | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 0.00% ±0.00 | 0.00% ±0.00 | 0.00 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 20.67% ±2.81 | 6.25% ±1.68 | 14.42 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 87.40% ±0.33 | 77.92% ±0.43 | 9.48 pts | Reported ±1 SE ranges do not overlap |
| EMB | 52.62% ±2.97 | 44.75% ±3.12 | 7.87 pts | Reported ±1 SE ranges do not overlap |
| Finance Agent (v2) | 42.98% ±1.21 | 38.22% ±1.19 | 4.77 pts | Reported ±1 SE ranges do not overlap |
| MortgageTax | 69.40% ±0.91 | 59.10% ±0.96 | 10.29 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 37.84% ±3.10 | 26.58% ±2.52 | 11.26 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 72.88% ±0.86 | 67.42% ±0.93 | 5.47 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 59.06% ±2.00 | 41.03% ±2.26 | 18.03 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 76.11% ±1.92 | 77.09% ±1.89 | 0.98 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 95.45% ±1.05 | 77.53% ±2.42 | 17.93 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 90.99% ±0.28 | 77.17% ±0.43 | 13.81 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 88.21% ±0.78 | 73.58% ±1.06 | 14.62 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 48.68% ±3.29 | 38.08% ±3.09 | 10.60 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 17.31% ±3.93 | 14.47% ±4.03 | 2.85 pts | Reported ±1 SE ranges overlap |
| LiveCodeBench | 88.48% ±0.93 | 84.01% ±1.04 | 4.48 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench | 78.80% ±1.83 | 69.80% ±2.06 | 9.00 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 32.03% ±4.34 | 26.10% ±5.08 | 5.94 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | Gemini 3.1 Pro Preview (02/26) average | GPT 5.4 Nano average |
|---|---|---|
| Legal | 36.02% | 28.06% |
| Finance | 55.14% | 47.21% |
| Healthcare | 67.59% | 59.06% |
| Academic | 91.55% | 76.09% |
| Education | 48.68% | 38.08% |
| Coding | 54.16% | 48.59% |
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 | Gemini 3.1 Pro Preview (02/26) cost | GPT 5.4 Nano cost | Gemini 3.1 Pro Preview (02/26) latency | GPT 5.4 Nano latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $1.15 | $0.18 | 7m17s | 6m03s |
| Legal Research Bench | $1.10 | $0.14 | 4m02s | 8m47s |
| LegalBench | N/A | N/A | 10.06s | 2.70s |
| EMB | $3.77 | $0.57 | 11m49s | 29m27s |
| Finance Agent (v2) | $1.59 | $0.16 | 3m28s | 5m35s |
| MortgageTax | N/A | N/A | 23.62s | 14.53s |
| Tax Agent Bench | $0.30 | $0.07 | 100.96s | 3m40s |
| TaxEval v2 | N/A | N/A | 41.93s | 15.90s |
| MedCode | N/A | N/A | 38.52s | 8.04s |
| MedScribe | N/A | N/A | 69.14s | 20.53s |
| GPQA Diamond | N/A | N/A | 65.76s | 29.50s |
| MMLU Pro | N/A | N/A | 23.91s | 8.21s |
| MMMU Pro | N/A | N/A | 76.99s | 21.91s |
| SAGE | N/A | N/A | 61.17s | 34.33s |
| Code Migration | $1.61 | $0.42 | 12m01s | 26m34s |
| LiveCodeBench | N/A | N/A | 88.89s | 60.09s |
| SWE-bench | $0.78 | $0.10 | 5m12s | 4m23s |
| Vibe Code Bench v1.1 | $3.83 | $1.28 | 20m12s | 54m36s |
Results available only for Gemini 3.1 Pro Preview (02/26)
- Vals Index
- ProofBench v1.1
- BioMysteryBench
- MysteryMechanism
- Terminal-Bench Science
- IOI
- ProgramBench
- Terminal-Bench 4.0
- Vibe Code Bench 1-100
- Public Benefits Bench v1.1
Results available only for GPT 5.4 Nano
None.