Gemini 3.8 Flash vs GPT-6 Astra: Benchmark Comparison
Gemini 3.8 Flash has the higher score on 5 of 23 shared benchmarks; GPT-6 Astra leads on 18.
The largest observed score gap is 71.67 pts on Time Horizon Index: KSP , where GPT-6 Astra leads.
Reported ±1 standard-error ranges overlap on 6 of 21 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 3.8 Flash | GPT-6 Astra | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 54.83% ±1.03 | 63.13% ±1.19 | 8.30 pts | Reported ±1 SE ranges do not overlap |
| Vals RSI Index | 19.93% | 27.07% | 7.14 pts | Uncertainty comparison unavailable |
| Harvey's Legal Agent Benchmark | 10.00% ±2.42 | 5.42% ±1.17 | 4.58 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 38.94% ±3.39 | 39.42% ±3.40 | 0.48 pts | Reported ±1 SE ranges overlap |
| EMB | 72.20% ±2.42 | 71.70% ±2.48 | 0.50 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 61.44% ±0.13 | 53.54% ±2.08 | 7.90 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 66.77% ±3.15 | 63.34% ±3.15 | 3.44 pts | Reported ±1 SE ranges overlap |
| MedCode | 48.13% ±2.18 | 48.49% ±2.13 | 0.35 pts | Reported ±1 SE ranges overlap |
| MedScribe | 84.50% ±1.94 | 87.91% ±1.94 | 3.41 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 48.00% ±5.02 | 99.00% ±1.00 | 51.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 62.22% ±1.70 | 79.26% ±0.98 | 17.04 pts | Reported ±1 SE ranges do not overlap |
| MysteryMechanism | 36.49% ±3.24 | 53.15% ±3.36 | 16.67 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench Science | 8.57% ±3.37 | 62.86% ±5.82 | 54.29 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 35.06% ±3.36 | 46.37% ±3.44 | 11.30 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 36.55% ±4.18 | 67.74% ±4.22 | 31.19 pts | Reported ±1 SE ranges do not overlap |
| IOI | 56.94% ±2.26 | 100.00% ±0.00 | 43.06 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 1.00% ±0.70 | 5.50% ±1.62 | 4.50 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 19.19% ±2.52 | 59.60% ±4.40 | 40.40 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 18.77% ±3.31 | 27.64% ±4.08 | 8.87 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 78.65% ±3.88 | 89.59% ±2.17 | 10.94 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 43.75% ±2.21 | 41.07% ±2.56 | 2.68 pts | Reported ±1 SE ranges overlap |
| CUA-bench | 4.17% | 19.17% | 15.00 pts | Uncertainty comparison unavailable |
| Time Horizon Index: KSP | 18.83% ±0.00 | 90.50% ±0.00 | 71.67 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Gemini 3.8 Flash average | GPT-6 Astra average |
|---|---|---|
| Legal | 24.47% | 22.42% |
| Finance | 66.80% | 62.86% |
| Healthcare | 66.32% | 68.20% |
| Math | 48.00% | 99.00% |
| Science | 35.76% | 65.09% |
| Education | 35.06% | 46.37% |
| Coding | 35.18% | 58.35% |
| Cyber | 43.75% | 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 | Gemini 3.8 Flash cost | GPT-6 Astra cost | Gemini 3.8 Flash latency | GPT-6 Astra latency |
|---|---|---|---|---|
| Vals Index | $5.73 | $18.46 | 51m57s | 27m48s |
| Vals RSI Index | $332.04 | $759.51 | 90h00m | 90h00m |
| Harvey's Legal Agent Benchmark | $3.66 | $26.16 | 29m21s | 24m49s |
| Legal Research Bench | $1.63 | $10.58 | 5m24s | 24m51s |
| EMB | $8.24 | $5.83 | 12m49s | 13m24s |
| Finance Agent (v2) | $2.00 | $6.82 | 3m21s | 11m33s |
| Tax Agent Bench | $0.86 | $5.80 | 2m58s | 15m01s |
| MedCode | N/A | N/A | 42.89s | 102.86s |
| MedScribe | N/A | N/A | 21.14s | 2m34s |
| ProofBench v1.1 | $0.60 | $1.71 | 6m09s | 5m08s |
| BioMysteryBench | $1.55 | $1.72 | 5m27s | 4m46s |
| MysteryMechanism | $1.73 | $1.56 | 5m22s | 8m09s |
| Terminal-Bench Science | $5.64 | $20.80 | 55m11s | 1h49m |
| SAGE | N/A | N/A | 28.64s | 41.63s |
| Code Migration | $18.49 | $44.36 | 2h37m | 58m30s |
| IOI | $3.98 | $6.50 | 14m20s | 31m40s |
| ProgramBench | $10.93 | $11.57 | 46m39s | 27m02s |
| Terminal-Bench 4.0 | $8.77 | $9.58 | 1h48m | 35m42s |
| Vibe Code Bench 1-100 | $18.91 | $84.72 | 1h06m | 3h26m |
| Vibe Code Bench v1.1 | $6.87 | $38.51 | 8m39s | 39m04s |
| CyberBench v1.1 | $1.31 | $1.84 | 9m49s | 4m10s |
| CUA-bench | $280.03 | $1821.37 | 36.55s | 32.09s |
| Time Horizon Index: KSP | $543.14 | $4206.59 | N/A | N/A |
Results available only for Gemini 3.8 Flash
- LegalBench
- MortgageTax
- TaxEval v2
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- LiveCodeBench
- SkillsBench
- SWE-bench
- Public Benefits Bench v1.1
Results available only for GPT-6 Astra
- SRE Bench