⚖️ Judge verdict
The AI Cold War's Newest Weapon: Stealing the Teacher's Answer Key · Judge: anthropic:claude-opus-4-8 · 2026-09-19T07:35:08
· ✅ saved (c2d9e3) · 3 models
🔬 View outputs side-by-side →
all verdicts →
🏆 Quality winner: openrouter:deepseek/deepseek-v4-pro
Both DeepSeek and Haiku are far richer than Ernie, offering full quizzes, deep glossaries and multi-paragraph passages. DeepSeek edges out Haiku with the most mechanistically precise analogy (Tu-4 reverse-engineering matches distillation's 'copy that doesn't match the original but is cheaper'), consistently self-contained passage prose, and a complete untruncated output, whereas Haiku's passage is cut off mid-sentence and a couple of its vocab questions test words not in its own rewritten passage.
💰 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 | 9 |
9 |
0.62¢ |
8.7 |
Excellent depth and accuracy; the Tu-4/B-29 Cold War analogy shares the reverse-engineering mechanism precisely. Ten well-constructed SAT-style questions including evidence-pairing and vocab-in-context, all answerable from the passage. Rich, original multi-paragraph passage and precise glossary. Minor: some questions reference the original FT text rather than only the rewritten passage. |
| 🥈 |
anthropic:claude-haiku-4-5-20251001
view ↗ |
9 | 9 | 9 | 9 | 8 | 9 | 9 |
8 |
2.57¢ |
6.8 |
Very strong: engaging, accurate, with sharp analogies (doping scandal, arms race) that fit the 'legal-vs-illicit' distinction well. Ten good questions, though Q4 ('surreptitious') and some evidence questions test the source text rather than the rewritten passage, slightly unfair. Passage is excellent and comprehensive; the truncated final word is a small flaw. |
| 🥉 |
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 clean structure and good analogies (recipe, plagiarism). But only 3 quiz questions, and Q2 answer 'replicate' over 'steal' is defensible but debatable; some quiz explanations reference 'traps' oddly. Passage is accurate but compresses heavily. |
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.