-
Konrad Przepiórzyński
Mingbird: When a Small Local Model Fails, the Harness Is Usually the One to Blame
You download an open-weight model onto a laptop – two…
-
Konrad Przepiórzyński
DeepSeek-V4.1-Flash: KV Cache Down to 890 Bytes per Token. 552B Parameters, 8B Active on Prefill
DeepSeek-V4.1-Flash’s technical report (arXiv 2609.19969, September 17, 2026) has a…
-
Konrad Przepiórzyński
DeepSeek Harness: Open-Source Agent Framework Where Everything Is a Plugin – 166K Stars in 3 Days
DeepSeek did it again. This time it’s not about a…
Latest Posts
- Mingbird: When a Small Local Model Fails, the Harness Is Usually the One to Blame
- DeepSeek-V4.1-Flash: KV Cache Down to 890 Bytes per Token. 552B Parameters, 8B Active on Prefill
- Anthropic Pushes Its IPO to November. It Wants $2 Trillion and a Full Quarter on the Table
- Enflame Debuts With a 179% Pop. The Chinese AI Chipmaker Is Worth $25 Billion and Has Never Had a Profitable Year
- DeepSeek Orders 160,000 Huawei Chips for $2.5B – The Largest AI Cluster Outside Nvidia