Welcome back to Agentic Coding Weekly. Here are the updates on agentic coding tools, models, and workflows for the week of Aug 16 - 22, 2026.
1. Tooling and Model Updates
Qwen 3.8 27B (xhigh) Analysis
27B variant was released about 10 days ago with strong benchmark results (matching Opus 4.6 max) for a model this size. Qwen 3.6 27B was the best model to run locally on a laptop for coding and 3.8 27B continues this legacy. Now the Artificial Analysis scores and first hand impression are available.
It scores 52 on AA Intelligence Index which is same score as GPT 5.6 Luna (max) and just one point lower than DeepSeek V4 Pro. It has the highest score in small (4B-40B) as well as medium models (40B-150B) category.
In terms of drawbacks, it tends to overthink on the default reasoning level, xhigh. Read the review of 4 bit quantized version from Simon Willison.
DeepSeek-V4-Flash-Vision-Exp
Deepseek V4 Flash is my favorite model these days and the missing vision capability is the only annoying thing about it. This new experimental version for the flash model now has the vision capability and it improves upon the coding benchmarks as well.
fx
Vercel released fx, an open-source coding agent harness written in Zig and focuses on minimalism and performance. Some distinguishing features are:
CLI output style and form factor aims to be closer to a Unix shell than a heavy "IDE in the terminal" TUI
cold starts in 10µs and does no unnecessary work or I/O prior to accepting user input, making it ideal for programmatic use
Ornith 1.5
Three model: 397B MoE, 35B MoE, and 9B Dense. It extends Ornith 1.0, which was developed on top of Qwen 3.5 and Gemma 4 with additional continued pre-training, mid-training, and post-training.
397B performs on par with Claude Opus 4.8 and outperforms open-weight models of similar scale, including GLM-5.2 and DeepSeek-V4-Flash-0731.
35B outperforms Qwen 3.6 35B on all coding and agentic benchmarks. Also outperforms dense models like Gemma 4 31B and Meta’s Muse Glimmer 30B.
Quick Updates
Codex CLI startup is really fast now (~25x faster).
Claude Code now has a built-in design skill.
Cursor launched Origin, a GitHub alternative. It's an early beta version.
debuzz is an agent skill that uses Antigravity CLI (agy) to make Claude stop talking like a BuzzFeed article.
2. Reading List
That’s it for this week. I’ll be back next Monday with the latest agentic coding updates.
— Prashant