2 vs 0 benchmarks won
Anthropic Claude 3.5 Haiku | DeepSeek DeepSeek V2.5 | |
|---|---|---|
| Overview | ||
| Company | Anthropic | DeepSeek |
| Release date | Oct 22 2024 | Sep 2024 |
| Model type | — | — |
| Open source | No | Yes |
| Specifications | ||
Parameters | — | — |
Context window | — | — |
| Benchmarks | ||
Science reasoning GPQA Diamond | 41.6% | — |
Software engineering SWE-Bench Verified | 40.6% | — |
Multimodal understanding MMMU | — | — |
| Timeline | ||
| Release gap | DeepSeek V2.5 shipped 51 days before Claude 3.5 Haiku | |
Claude 3.5 Haiku leads DeepSeek V2.5 on 2 of the tracked benchmarks (GPQA Diamond, SWE-Bench Verified, MMMU). DeepSeek V2.5 shipped 51 days before Claude 3.5 Haiku, so benchmark comparisons should account for the intervening progress.
DeepSeek V2.5 is an open-source / open-weight model; Claude 3.5 Haiku is proprietary.
Direct benchmark comparisons are unavailable — at least one of these models has not published scores on GPQA Diamond, SWE-Bench Verified, or MMMU.