Compare / Claude Haiku 4.5 vs DeepSeek V4 Pro 0423
Claude Haiku 4.5 vs DeepSeek V4 Pro 0423
21 results from 4 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Claude Haiku 4.5 | DeepSeek V4 Pro 0423 |
|---|---|---|---|
| ROASBenchmeasured by us | Audience scorescore out of 100 | 30.9 | 44.8 |
| ROASBenchmeasured by us | Business scorescore out of 100 | 7.3 | 23.8 |
| ROASBenchmeasured by us | Consistency scorescore out of 100 | 8.7 | 9.1 |
| ROASBenchmeasured by us | Contribution profitsimulated US dollars | −$205,001 | $301,289 |
| ROASBenchmeasured by us | Cost of a runUS dollars | $0.0298 | $0.20 |
| ROASBenchmeasured by us | Months over budgetmonths of 12 | 2 | 0 |
| ROASBenchmeasured by us | Overall scorescore out of 100 | 17.5 | 28.2 |
| ROASBenchmeasured by us | Planning scorescore out of 100 | 54.4 | 54.0 |
| ROASBenchmeasured by us | Return on ad spendprofit per $1 spent | −$0.21 | $0.26 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,418 | 1,455 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,420 | 1,460 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,387 | 1,446 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,452 | 1,480 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,415 | 1,452 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,414 | 1,458 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,399 | 1,451 |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 6.0% | 34.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.39 | 0.36 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 52.2 | 68.4 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 99.5% | 97.2% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 9.8% | 8.6% |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.