⚖️ Judge verdict
The AI Cold War's Newest Weapon: Stealing the Teacher's Answer Key · Judge: anthropic:claude-opus-4-8 · 2026-09-25T04:35:04
· ✅ saved (7ec8b6) · 3 models
🔬 View outputs side-by-side →
all verdicts →
🏆 Quality winner: openrouter:deepseek/deepseek-v4-pro
2eb734 edges out 0313fc on the strength of its complete, polished reading passage and its distinctive mechanism-matching analogies (Tu-4 bomber, forking code), plus its clean, non-truncated output. Both are far richer than a0132a, which has only three questions and less depth. 2eb734's quiz is fully self-contained and its vocab questions ('frontier', 'distillation') are grounded in its own passage, whereas 0313fc leans on 'surreptitious', a word it dropped from the rewrite.
💰 Best value: openrouter:deepseek/deepseek-v4-pro
— quality 9/10 at 0.62¢ → value 8.7/10
Value = 0.7·quality + 0.3·cheapness. Quality stays dominant, so a cheap-but-weak output can't win on price alone.
Quality ranking
| # | Model | Facts | Analogy | Age-fit | Clarity | Quiz | Glossary | Passage | Overall | Cost | Value | Notes |
|---|
| 🥇 |
openrouter:deepseek/deepseek-v4-pro
view ↗ |
9 | 9 | 9 | 9 | 9 | 9 | 10 |
9 |
0.62¢ |
8.7 |
Excellent, richly detailed. The Tu-4 bomber and term-paper analogies genuinely share the mechanism of reverse-engineering/extracting knowledge. Ten well-constructed SAT-style questions including proper evidence-pairing, precise glossary, and an outstanding original reading passage. Minor: dates slightly reworked (Feb 2026 vs source's ambiguous 'February') but reasonable. |
| 🥈 |
anthropic:claude-haiku-4-5-20251001
view ↗ |
9 | 9 | 9 | 9 | 8 | 9 | 9 |
8 |
2.57¢ |
6.8 |
Very strong: doping-scandal and arms-race analogies fit the mechanism well, six glossary terms are precise, and ten quiz questions include good evidence-pairing. One 'surreptitious' vocab question relies on a word in the source not carried into the rewritten passage. Reading passage appears truncated at the end but is otherwise excellent. |
| 🥉 |
openrouter:baidu/ernie-4.5-vl-424b-a47b
view ↗ |
8 | 7 | 8 | 8 | 6 | 8 | 8 |
7 |
0.43¢ |
7.9 |
Solid, faithful summary with good chef/recipe hook and clean glossary. Only 3 quiz questions, and the 'distil' vocab item is debatable (answer C 'replicate' vs D 'steal' is not clearly grounded). Fewer questions and shorter than competitors, but accurate throughout. |
Scores 0–10 · 🟩 ≥8 · 🟨 4–7 · 🟥 <4 · sorted by Overall. Hover a column header for its full rubric name.
💰 Cost-adjusted ranking (value for money)
| # | Model | Quality | Cost | Value |
|---|
| 🥇 |
openrouter:deepseek/deepseek-v4-pro |
9 |
0.62¢ |
8.7 |
| 🥈 |
openrouter:baidu/ernie-4.5-vl-424b-a47b |
7 |
0.43¢ |
7.9 |
| 🥉 |
anthropic:claude-haiku-4-5-20251001 |
8 |
2.57¢ |
6.8 |
🪙 Best under 3¢ (cost < $0.03), ranked by quality
| # | Model | Quality | Cost | Value |
|---|
| 1 |
openrouter:deepseek/deepseek-v4-pro |
9 |
0.62¢ |
8.7 |
| 2 |
anthropic:claude-haiku-4-5-20251001 |
8 |
2.57¢ |
6.8 |
| 3 |
openrouter:baidu/ernie-4.5-vl-424b-a47b |
7 |
0.43¢ |
7.9 |
Strong quality without paying flagship prices — the cheap-and-good picks.