Note: the judge model is also one of the contestants
— scores may carry self-preference bias.
🏆 Quality winner: openrouter:deepseek/deepseek-v4-pro
DeepSeek-v4 wins on the strongest combination of faithful fact-grounding (no invented statistics, unlike Sonnet's fabricated 16M/24,000 figures), a genuinely mechanism-matching Cold War Tu-4/B-29 analogy that captures reverse-engineering-that-closes-a-gap-but-never-matches, a complete and engaging SAT-style passage, and a full, fair 10-question quiz whose items are answerable from the text with strong evidence-pairing. Claude Opus (c2b157) and Sonnet (fe7a5d) are very close, but Opus's Q4/Q5 lean on its own passage vocabulary and Sonnet invents specifics; DeepSeek avoids both pitfalls while matching their polish.
| # | Model | Facts | Analogy | Age-fit | Clarity | Quiz | Glossary | Passage | Overall | Cost | Value | Notes |
|---|
| 🥇 |
anthropic:claude-sonnet-4-6
view ↗ |
8 | 9 | 9 | 9 | 9 | 9 | 8 |
9 |
8.57¢ |
6.6 |
Excellent analogies (chef ordering every dish, band recordings) that match the mechanism precisely. Introduces some invented specifics (16M exchanges, 24,000 accounts) not in the source, a minor grounding risk. Strong 10-question quiz with good SAT tips; passage truncated but high quality. |
| 🥈 |
openrouter:deepseek/deepseek-v4-pro
view ↗ |
9 | 9 | 9 | 9 | 9 | 8 | 9 |
9 |
0.62¢ |
8.7 |
Very faithful; strong Cold War Tu-4/B-29 analogy sharing the reverse-engineering mechanism. Complete, engaging rewritten passage. Full 10-question quiz with tone and evidence-pairing items well grounded. |
| 🥉 |
openrouter:minimax/minimax-m2.7
view ↗ |
9 | 8 | 9 | 9 | 9 | 9 | 9 |
9 |
0.62¢ |
8.7 |
Highly faithful and clear; university/answer-key and F1-footage analogies fit the mechanism well. Complete, engaging passage; strong quiz with good evidence pairing. One of the best-balanced entries. |
| 4 |
openrouter:mistralai/mistral-large-2512
view ↗ |
8 | 7 | 8 | 8 | 8 | 8 | 8 |
8 |
0.85¢ |
8.0 |
Solid, well-structured explainer with a complete passage. Marathon/save-scumming analogies are decent but the 'save scumming' one drifts from the actual mechanism. Facts largely accurate; quiz is fair and grounded. |
| 5 |
openrouter:moonshotai/kimi-k2.6
view ↗ |
8 | 9 | 9 | 9 | 8 | 8 | 9 |
8 |
9.62¢ |
5.9 |
Sophisticated conservatory/pharma analogies matching mechanism; elegant complete passage. Quiz Q4/Q5 ask about words ('sharp', 'lightweight') that appear only in its own rewritten passage rather than the source, a mild fairness issue but internally consistent. |
| 6 |
openrouter:google/gemini-3.1-flash-lite
view ↗ |
8 | 7 | 8 | 8 | 8 | 8 | 8 |
8 |
0.59¢ |
8.0 |
Chef-tasting analogy fits; clean complete passage and a full, fair 10-item quiz with good tone/evidence questions. Solid all-around, slightly less distinctive than the top entries. |
| 7 |
openrouter:z-ai/glm-5-turbo
view ↗ |
7 | 8 | 9 | 9 | 8 | 8 | 8 |
8 |
3.40¢ |
6.8 |
Vivid music-producer analogy and lively 'Why It Matters'. Q4 asks about 'sharp' which appears only in its own passage; adds McGuire details fabricated slightly ('billions', 'from scratch'). Passage truncated but strong. |
| 8 |
openrouter:openai/gpt-5.4-nano
view ↗ |
8 | 7 | 8 | 8 | 8 | 8 | 8 |
8 |
0.63¢ |
8.0 |
Careful, faithful GPT-5-style entry that explicitly attributes claims to 'the article', reducing overclaiming. Complete passage and a well-constructed, grounded 10-item quiz. Analogies (locked lab, film scouting) are good if slightly generic. |
| 9 |
anthropic:claude-haiku-4-5-20251001
view ↗ |
8 | 7 | 8 | 8 | 8 | 8 | 8 |
8 |
2.57¢ |
6.8 |
Strong nuclear-arms-race and doping analogies (doping is apt for the legal/illegal distinction). Complete passage, full grounded quiz with excellent evidence-pairing. Truncated final paragraph is the only blemish. |
| 10 |
openrouter:stepfun/step-3.7-flash
view ↗ |
7 | 8 | 8 | 8 | 8 | 8 | 7 |
7 |
2.09¢ |
6.1 |
Good SAT/burger analogies and thorough quiz with trap-labeling. Adds unsupported specifics ('state-linked groups', memo 'names three firms', bills 'passed April 2026') that overstate the source. Passage truncated. |
| 11 |
openrouter:x-ai/grok-4.3
view ↗ |
7 | 6 | 8 | 8 | 8 | 8 | 8 |
7 |
1.64¢ |
6.7 |
Clear and neutral with reasonable scouting/reverse-engineering analogies. Complete passage and full quiz. Some quiz items reference details ('$1 trillion' absent) or added framing; grounding adequate but not exceptional. |
| 12 |
openrouter:qwen/qwen3-max-thinking
view ↗ |
8 | 8 | 8 | 7 | 5 | 8 | 8 |
6 |
1.02¢ |
6.0 |
Concise and accurate with good handbag/answer-key analogies and a clean passage. Major weakness: only ONE quiz question provided, well below the expected set, sharply limiting quiz value. |
| 13 |
openrouter:amazon/nova-pro-v1
view ↗ |
7 | 5 | 6 | 6 | 6 | 6 | 6 |
6 |
1.34¢ |
6.0 |
Generic chess/doping analogies that don't capture the distillation mechanism well. Q9/Q10 rely on a line ('significant implications for global power dynamics') that exists only in its own added text, weakening grounding. Serviceable but thin. |
| 14 |
openrouter:bytedance-seed/seed-2.0-lite
view ↗ |
6 | 7 | 8 | 8 | 6 | 8 | 7 |
6 |
1.73¢ |
6.0 |
Q4 asks meaning of 'principal' claiming the passage describes 'Chinese entities' as 'principal'—but neither its passage nor source uses that adjective that way, an unfair/fabricated item. Otherwise decent MIT-lab analogy and complete passage. |
| 15 |
openrouter:baidu/ernie-4.5-vl-424b-a47b
view ↗ |
7 | 6 | 7 | 7 | 5 | 6 | 7 |
6 |
0.43¢ |
7.2 |
Compact and readable with a chef-recipe hook, but only THREE quiz questions and three glossary terms, well short of expectations. Analogies are somewhat generic; passage complete but brief. |
Scores 0–10 · 🟩 ≥8 · 🟨 4–7 · 🟥 <4 · sorted by Overall. Hover a column header for its full rubric name.
| # | Model | Quality | Cost | Value |
|---|
| 🥇 |
openrouter:deepseek/deepseek-v4-pro |
9 |
0.62¢ |
8.7 |
| 🥈 |
openrouter:minimax/minimax-m2.7 |
9 |
0.62¢ |
8.7 |
| 🥉 |
openrouter:mistralai/mistral-large-2512 |
8 |
0.85¢ |
8.0 |
| 4 |
openrouter:google/gemini-3.1-flash-lite |
8 |
0.59¢ |
8.0 |
| 5 |
openrouter:openai/gpt-5.4-nano |
8 |
0.63¢ |
8.0 |
| 6 |
openrouter:baidu/ernie-4.5-vl-424b-a47b |
6 |
0.43¢ |
7.2 |
| 7 |
openrouter:z-ai/glm-5-turbo |
8 |
3.40¢ |
6.8 |
| 8 |
anthropic:claude-haiku-4-5-20251001 |
8 |
2.57¢ |
6.8 |
| 9 |
openrouter:x-ai/grok-4.3 |
7 |
1.64¢ |
6.7 |
| 10 |
anthropic:claude-sonnet-4-6 |
9 |
8.57¢ |
6.6 |
| 11 |
openrouter:stepfun/step-3.7-flash |
7 |
2.09¢ |
6.1 |
| 12 |
openrouter:qwen/qwen3-max-thinking |
6 |
1.02¢ |
6.0 |
| 13 |
openrouter:amazon/nova-pro-v1 |
6 |
1.34¢ |
6.0 |
| 14 |
openrouter:bytedance-seed/seed-2.0-lite |
6 |
1.73¢ |
6.0 |
| 15 |
openrouter:moonshotai/kimi-k2.6 |
8 |
9.62¢ |
5.9 |
| # | Model | Quality | Cost | Value |
|---|
| 1 |
openrouter:deepseek/deepseek-v4-pro |
9 |
0.62¢ |
8.7 |
| 2 |
openrouter:minimax/minimax-m2.7 |
9 |
0.62¢ |
8.7 |
| 3 |
openrouter:mistralai/mistral-large-2512 |
8 |
0.85¢ |
8.0 |
| 4 |
openrouter:google/gemini-3.1-flash-lite |
8 |
0.59¢ |
8.0 |
| 5 |
openrouter:openai/gpt-5.4-nano |
8 |
0.63¢ |
8.0 |
| 6 |
anthropic:claude-haiku-4-5-20251001 |
8 |
2.57¢ |
6.8 |
| 7 |
openrouter:x-ai/grok-4.3 |
7 |
1.64¢ |
6.7 |
| 8 |
openrouter:stepfun/step-3.7-flash |
7 |
2.09¢ |
6.1 |
| 9 |
openrouter:baidu/ernie-4.5-vl-424b-a47b |
6 |
0.43¢ |
7.2 |
| 10 |
openrouter:qwen/qwen3-max-thinking |
6 |
1.02¢ |
6.0 |
| 11 |
openrouter:amazon/nova-pro-v1 |
6 |
1.34¢ |
6.0 |
| 12 |
openrouter:bytedance-seed/seed-2.0-lite |
6 |
1.73¢ |
6.0 |
Strong quality without paying flagship prices — the cheap-and-good picks.