DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding Gains

Article automatically generated from technical news.

The architecture is identical to the April preview. Same 284B total parameters. Every gain came from re-post-training. 1. It beats the bigger model in its own family → Terminal Bench 2.1: 82.7 vs 72.1 for V4-Pro (Preview) → DeepSWE: 54.4 vs 12.8 → Toolathlon-Verified: 70.3 vs 55.9 → NL2Repo: 54.2 vs 38.5 V4-Pro carries 1.6T parameters with 49B activated. The smaller model wins on every agentic benchmark DeepSeek published. 2. The pricing gap is the story → $0.14 per 1M input tok

Fonte originale

DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding Gains

DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding Gains

Related Articles

New tool release: I built a security and governance layer to stop AI agents from leaking data to external APIs.

Multi-Agent Chat: why one "colossal agent" holding all your memory is the wrong default

Claude published malicious code to the Internet and attacked 3 real companies

Everyone is building LLM routers, we deprecated ours

With release of Deepseek V4 I wanted see how the model sizes are trending over time. Open source models are constantly getting smaller and better. The trend is that by this time next year, we probably will have Opus 4.5 level models on consumer grade laptops (sounds unlikely?!).