| Result | Rank 1GPT-6 Luna Low | Rank 2GPT-6 Sol Low | Rank 3Claude Haiku 5.5 Medium | Rank 4Grok 4.6 Low | Excluded: availability-failedQwen 3.8 27b Heretic Mlx Medium |
|---|---|---|---|---|---|
| Combined score | 89.51 | 86.12 | 74.03 | 63.71 | N/A |
| Judge | |||||
| Judge score /100 | 90.00 | 90.00 | 76.50 | 66.50 | N/A |
| Correctness /10 | 9 | 9 | 7 | 6 | N/A |
| Coverage /10 | 9 | 9 | 9 | 8 | N/A |
| Evidence /10 | 9 | 9 | 7 | 6 | N/A |
| Actionability /10 | 9 | 9 | 8 | 7 | N/A |
| Time | 11.94 s | 17.37 s | 61.28 s | 46.03 s | N/A |
| Estimated USD | $0.0015266 | $0.03029 | $0.0085056 | $0.038494 | N/A |
Cost estimates are reference estimates, not billed charges.
Frozen reference rate sheet: openrouter-rate-sheet-v1:d136d7bf3bdd121396516e965bdbddfb26414ec8725ca8c8bdee8ed2618e409c. API-equivalent estimate; subscription quota and charges are unknown.
Judge score uses recorded rubric metrics; final rank uses the combined score.
Component points and token usage
| Result | Rank 1 GPT-6 Luna Low | Rank 2 GPT-6 Sol Low | Rank 3 Claude Haiku 5.5 Medium | Rank 4 Grok 4.6 Low | Rank N/A Qwen 3.8 27b Heretic Mlx Medium |
|---|---|---|---|---|---|
| Quality points | 81 | 81 | 68.85 | 59.85 | N/A |
| Test points | 0 | 0 | 0 | 0 | N/A |
| Speed points | 4.1703 | 3.8774 | 2.4736 | 2.8293 | N/A |
| Cost points | 4.3378 | 1.241 | 2.7019 | 1.0311 | N/A |
| Total input incl cache | 9,321 | 9,320 | 16,236 | 10,103 | N/A |
| Cache read (subset of input) | 0 | 0 | 0 | 512 | N/A |
| Cache write (subset of input) | 0 | 0 | 16,234 | 0 | N/A |
| Total output incl reasoning | 1,189 | 1,165 | 13,764 | 3,048 | N/A |
| Reasoning (subset of output) | N/A | N/A | N/A | N/A | N/A |
Swipe, scroll or use Previous and More to compare configurations. Row labels stay visible.
Advanced results and scoring details
data · 2026-10-14T05:36:29.133Z · Prospective evaluation. Equal transport and native system conditions are unverified.
Applied weights: quality 90%, tests 0%, speed 5%, cost 5%.
Displayed scores are rounded; pairwise outcomes use exact recorded fractions. Reference costs are estimates, not billed charges. Missing components follow the recorded assessment.
Pairwise results
- Claude Haiku 5.5 Medium vs Grok 4.6 Low: First configuration wins
- Claude Haiku 5.5 Medium vs GPT-6 Luna Low: Second configuration wins
- Claude Haiku 5.5 Medium vs GPT-6 Sol Low: Second configuration wins
- Grok 4.6 Low vs GPT-6 Luna Low: Second configuration wins
- Grok 4.6 Low vs GPT-6 Sol Low: Second configuration wins
- GPT-6 Luna Low vs GPT-6 Sol Low: First configuration wins