Veladan Programming Blog

I posted this on X after finishing two full runs of DeepSWE, and it turned into the most interesting question of the whole exercise, so here is the long version.

This is the companion run to the GLM-5.3-Flash evaluation: same 113 tasks, same agent framework, same fixed task order, same binary grading.

I ran the whole DeepSWE v1.1 corpus against GLM-5.3-Flash, quantized to four bits and served locally on two consumer Blackwell cards.

DeepSWE covers five languages, and Rust is the smallest slice of it: five tasks out of 113.

Everyone building agent tooling shares a quiet hope: that enough instructions, skills and custom tools can compensate for a weaker model.

My whole career I’ve carried a balance that seemed permanent: scripting languages (Ruby, Python, JavaScript) buy you iteration speed and pay for it with errors that only surface at runtime; compiled languages (C, Java, Rust) buy you guarantees and pay with

A local voice assistant faces a small, annoying asymmetry on every single turn.

Seneschal is a real-time voice assistant: a single Rust binary where a microphone, a VAD, Whisper, an LLM streaming hundreds of tokens per second, a sentence splitter, and a TTS engine all have to cooperate within a few hundred milliseconds.

Every voice assistant demo ends with someone saying “it’s so fast.”

The classic voice-assistant architecture has a seductive failure mode.

Part one of this series covered quality, quantization, and the 1M-context price tag.

Qwen3.8 is out, and it’s a dense 27B that Qwen positions as the most capable generation in its open-model family.

In the current Artificial Intelligence ecosystem, the label “Open Source” has become an elastic, almost promotional term.

There are two dominant ways people use AI to develop software right now.

A recent debate regarding forest fires in Spain sparked a philosophical realization about the nature of freedom in the modern state.

I’ve been testing several high-performance models on the HermesAgent-20 benchmark suite via BenchLocal.app.

Today I ran a series of local benchmarks comparing poolside/Laguna-S-2.1-NVFP4 against Qwen3.6-27B-FP8 on my RTX PRO 6000 Blackwell.

I had my multi-agent cost strategy backwards.

I replaced my local model in OpenCode from Qwen3.6-27B to Qwen3.6-35B-A3B in FP8.

Why the Modular Pipeline is the Engine of Production Voice AI In the rapidly evolving landscape of Speech-to-Speech (S2S) models, the temptation to move toward “native” multimodal models is strong.