DeepSeek-Terminus
DeepSeek·
LLMsopen-weightFrontier
Overview
Experimental model pushing the boundaries of open weights. Achieved strong reasoning and coding scores rivaling closed models.
Capabilities and innovations
Advanced reasoning capabilitiesStrong coding performance128K context windowOpen-weight frontier modelNext-generation reasoning architectureImproved reward modelingEnhanced self-verification
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 83.7% | unsourced |
| Humanity’s Last Exam (no tools) | 29.5% | unsourced |
| SWE-bench Verified | 58.3% | unsourced |
| MMLU | 89.1% | unsourced |
Architecture and hardware
- Parameters
- 671B total, 37B active per token (MoE)
- Estimated VRAM at Q4
- ~386 GB, Frontier class
- Quantization formats
- GGUF, GPTQ, FP8
- Recommended runtime
- vLLM
- License
- MIT
Links
More from DeepSeek
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.