[ Category ]

Open Weights & Code LLMs

All technical deep dives, system architectures, and AI breakdowns, ordered by date.

5articles published

All categories Agent Systems & APIs (23)Vision & Multimodal LLMs (5)Open Weights & Code LLMs (5)AI Economics & Pricing (3)AI Safety & Security (13)Browser Agents (2)AI Hardware & Silicon (3)Manifesto & Systems (1)Self-Hosting & Open Source (36)

Previous Publications

Sep 10, 2026•9 min read

DeepSeek Launches V4.1-Flash: A Production Causal Encoder-Decoder, Native Vision, and a 4x Smaller KV Cache

DeepSeek released V4.1-Flash today, the smallest model in its new architecture family: a 552B MoE built on a Causal Encoder-Decoder that activates 8B parameters for input and 16B for output, an 890-byte-per-token KV cache, vision trained in from the start, 1M-token context, and lower API pricing with a plan to route all V4-Pro traffic to it. Weights are open under MIT, and the official comparison table shows exactly where it leads Opus 5 and where it trails.

Read post→

Open Weights & Code LLMs

Aug 28, 2026•5 min read

Tencent Ships Hy4 Preview: A 770B-Parameter Apache 2.0 Challenge to the Open-Weights Order

Tencent's Hy team released Hy4 preview — 770B total, 49B active, 1M-token context, full weights under Apache 2.0 on Hugging Face with an FP8 variant and official vLLM/SGLang recipes. A preview with real limitations, but the most permissive flagship license in the open-field race so far.

Read post→

Open Weights & Code LLMs

Aug 28, 2026•6 min read

GLM-5.3 Open Weights Are Live: Z.ai Ships Its 753B-Parameter Coding Frontier — With a Security-Review Clause

Two weeks after launch, Z.ai published the full GLM-5.3 weights on Hugging Face: 753B parameters in FP8, a 1M-token context, and a custom license that requires a security review before giant MaaS providers can use it commercially. Here is what is actually downloadable, and what it means.

Read post→

Open Weights & Code LLMs

Aug 28, 2026•6 min read

Qwen3.8-Flash-Next: Alibaba Opens the Architecture Behind Qwen4

Alibaba's Qwen team released Qwen3.8-Flash-Next as open weights on Hugging Face — a 125B-parameter MoE model with only 6B active per token, redesigned across attention, residuals, embeddings, and optimizer to preview the architecture behind the upcoming Qwen4 family.

Read post→

Open Weights & Code LLMs

Aug 27, 2026•5 min read

GLM-5.3-Flash: Near-Opus 4.8 Coding Performance at a Tenth of the Cost — Weights on Hugging Face, MIT Licensed

Z.ai's GLM-5.3-Flash is the first natively multimodal GLM-5 model: 320B total parameters, 18B active, a 1M-token context window, and MIT-licensed weights already on Hugging Face.

Read post→