FLUX.2 [dev]
Run 1
![FLUX.2 [dev] Run 1](/_next/image?url=https%3A%2F%2Fpub-568d40d307bf46efb39d070a9ca6a9ee.r2.dev%2Fruns%2Fcmrq4f3540009x7p8x0k2ota2%2Fthumb.webp&w=3840&q=75)
4.0
$0.012
Run 2
![FLUX.2 [dev] Run 2](/_next/image?url=https%3A%2F%2Fpub-568d40d307bf46efb39d070a9ca6a9ee.r2.dev%2Fruns%2Fcmrq4f672000bx7p88y1dir97%2Fthumb.webp&w=3840&q=75)
3.5
$0.012
Run 3
![FLUX.2 [dev] Run 3](/_next/image?url=https%3A%2F%2Fpub-568d40d307bf46efb39d070a9ca6a9ee.r2.dev%2Fruns%2Fcmrq4f977000dx7p8uy70i2k8%2Fthumb.webp&w=3840&q=75)
4.0
$0.012
Pass@1
Yes
Pass@3
Yes
Usable Rate
67%
Cost / Usable
$0.018
3-Run BenchmarkUchiwa T2I Benchmark 2026methodology v1.0
同一プロンプト・同一目標アスペクト比・各モデルの標準品質設定で3回ずつ生成した結果を比較しています。Pass@1 は初回生成が使用可能だったか、Pass@3 は3回以内に1件以上使用可能な生成があったかを表します。
| 順位 | モデル | Pass@1 | Pass@3 | Usable Rate | Avg Score | Total Cost | Cost / Usable |
|---|---|---|---|---|---|---|---|
| 1 | FLUX.2 [dev] | Yes | Yes | 67% | 3.8 | $0.036 | $0.018 |
| 2 | NanoBanana 2 | Yes | Yes | 67% | 3.8 | $0.240 | $0.120 |
| 3 | GPT Image 2 | No | Yes | 67% | 3.5 | $0.240 | $0.120 |
3試行ずつの生成結果
NanoBanana 2
Run 1
4.0
$0.080
Run 2
4.0
$0.080
Run 3
3.5
$0.080
Pass@1
Yes
Pass@3
Yes
Usable Rate
67%
Cost / Usable
$0.120
GPT Image 2
Run 1
2.5
$0.080
Run 2
4.0
$0.080
Run 3
4.0
$0.080
Pass@1
No
Pass@3
Yes
Usable Rate
67%
Cost / Usable
$0.120