Deepseek-V3-Base was just opensourced
- 685B MoE w/ 256 experts topk=8 with sigmoid routing
- Outperforms Sonnet 3.5 on Aider benchmark
The "preview" of DeepSeek's new V3 model now ranks 1st on BigCodeBench-Hard.
Complete -40.5%
Instruct -28.4%
Average -34.5%
Gemini-Exp-1206 Average -34.1%
o1-2024-12-17 (reasoning=medium) Average -32.8%.
This was benchmarked against the normal DeepSeek API. Apparently it has been upgraded from V2.5 to V3 Preview.
- 685B MoE w/ 256 experts topk=8 with sigmoid routing
- Outperforms Sonnet 3.5 on Aider benchmark
The "preview" of DeepSeek's new V3 model now ranks 1st on BigCodeBench-Hard.
Complete -40.5%
Instruct -28.4%
Average -34.5%
Gemini-Exp-1206 Average -34.1%
o1-2024-12-17 (reasoning=medium) Average -32.8%.
This was benchmarked against the normal DeepSeek API. Apparently it has been upgraded from V2.5 to V3 Preview.