DeepSeek V4 Pro hits 80.6% on SWE-Bench — open source #Shorts

DeepSeek just released V4 Pro — a 1.6 trillion parameter open-weights model that scores 80.6% on SWE-Bench Verified. Only 49 billion params are active per token thanks to mixture-of-experts. Released April 24, 2026.

Why this matters:
- Open model now matches the closed frontier on real coding benchmarks
- 1.6T total / 49B active MoE design
- 1 million token context window
- Available on HuggingFace, free to download and fine-tune

The gap between open and closed AI is closing fast.

Source: https://api-docs.deepseek.com/news/news260424

Is open source AI already winning the race? Comment below.

#Shorts #DeepSeek #DeepSeekV4 #AInews #AItools #LLM #AIcoding #SWEBench #openweights #ainews2026

Видео DeepSeek V4 Pro hits 80.6% on SWE-Bench — open source #Shorts канала ProCode