thirty first July 2026 – Hyperlink Weblog
deepseek-ai/DeepSeek-V4-Flash-0731 (by way of) The most recent launch in DeepSeek’s V4 household, “with considerably enhanced agentic capabilities”. It is 304 billion parameters – 167GB on Hugging Face – however it seems to punch effectively above its weight.
Synthetic Evaluation rank it forward of MiniMax M3 – a 428B mannequin. It is $0.14/million enter and $0.27/million output pricing means this may increasingly at the moment be the very best value-per-intelligence mannequin on the market. It is wanting superb on the Intelligence Index vs. Value per Intelligence Index Job chart:
I acquired a disappointing pelican from it utilizing the default reasoning stage by way of OpenRouter:
However once I bumped reasoning stage as much as excessive I acquired one thing significantly better:
llm -m openrouter/deepseek/deepseek-v4-flash-0731 -t pelican -o reasoning_effort excessive
