The Ghost in the Weights

Unemployment Labs published a forensic analysis showing that six major Chinese AI models — including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3 — carry statistical signatures in their weights that match Nvidia's Blackwell GPU architecture, which is currently banned from export to China. The method is technically elegant, legally limited, and politically timely.

Read more →

284 Billion Parameters, 64 Gigabytes of RAM

Synchopate's llama-modchip runs DeepSeek-V4-Flash (284B MoE) on a CPU-only 64GB DDR5 machine at 9–14.7 tok/s — no GPU required. The engineering is real but the memory margin is razor-thin, the format is non-standard, and the interesting question isn't whether it fits but what you'd actually use it for.

Read more →

DeepSeek Opens the Inference Stack

DeepSeek has released DeepInfer, the inference framework behind its production services, connecting FlashMLA, DeepGEMM, DeepEP, and TileLang into a single stack under Apache 2.0. Cross-hardware support for NVIDIA SM100 and Huawei Ascend 950 from day one makes this more than just another vLLM fork.

Read more →

DeepSeek Tunes Flash for the Agentic Stack

DeepSeek quietly dropped a revision of V4.1 Flash on October 17th. The benchmarks call it a "minor update" but the jumps on Terminal-Bench and ProgramBench aren't minor — they reflect targeted work on the agentic task distribution that practitioners actually run.

Read more →

The Sandbox Numbers

DeepSeek's DSec paper describes the infrastructure behind their agentic training: 160 nodes, 380k concurrent sandboxes, 5,000+ creations per second, roughly 3 million environments per day. The architecture choices — layered images, memory sharing, typed sandbox registry, separation of stateful rollouts from preemptible GPU jobs — solve a problem most labs are only beginning to hit.

Read more →

DeepSeek's Plugin-First Agent Runtime

DeepSeek released Harness v0.1, an MIT-licensed open-source agent framework where every component — model, tools, storage, agent loop, UI — is a swappable plugin on a Cordis microkernel. Released alongside a Peking University research paper on the formal foundations of dynamic component composition.

Read more →

Same Weights, Different Agent

DeepSeek V4-Flash-0731 went from a 7.3 to a 54.4 on DeepSWE with identical pretrained weights — a 645% jump achieved purely through post-training. It's evidence that the gap between "can write code" and "can act as an agent" is largely a training-envelope problem, not a capacity problem, and that post-training is now a first-class axis of model improvement alongside pretraining scale and architecture.

Read more →

DeepSeek Ships Speculative Decoding to Production and Open-Sources the Whole Stack

DeepSeek released DSpark on June 27 — a semi-parallel speculative decoding framework already running in production for DeepSeek-V4 — alongside DeepSpec, an MIT-licensed toolkit packaging three drafting algorithms with complete training and evaluation pipelines. Together they let anyone train a custom draft model for their own target LLM, not just the models DeepSeek ships.

Read more →

AMD's FP8 Problem, and What It Costs

A detailed engineering account of bringing DeepSeek-V4-Flash up on AMD MI300X reveals the real cost of AMD's software ecosystem gaps: FP8 format fragmentation, missing kernels, and HIP graph constraints that each required dedicated engineering effort before getting to 2,700 tokens/s.

Read more →

The Terminal Agent That Bets Everything on the Cache

DeepSeek Reasonix is a DeepSeek-native terminal coding agent that treats prefix-cache stability as a first-class invariant rather than a side effect. With 99.82% cache hit rates in reported benchmarks, it cuts a heavy session from ~$61 to ~$12 — deliberately by coupling tightly to one provider's caching behavior instead of staying provider-agnostic.

Read more →

One Model, One Chip, No Framework

Salvatore Sanfilippo (antirez, Redis) released ds4: a single-model Metal inference engine for DeepSeek V4 Flash that deliberately rejects the general-framework approach. Asymmetric 2-bit quantization on MoE experts only gets a 280B-parameter model into 128 GB RAM with 26–36 t/s generation, 1M-token context, and disk-persisted KV cache on Apple Silicon.

Read more →