AUTHOR · FOUNDER & EDITOR

Muhammad Kashif

Founder and editor of Devventa. Writes about AI coding assistants, Next.js, and the workflows behind shipping production software with AI tools.

LinkedIn GitHub Facebook

Articles by Muhammad · 104

SWE-bench Verified counted by repository, with Django taking almost half the instances

Guides

Best AI Model for Python: 46% of the Evidence Is Django

Aug 19 · 12 min

The same model answering one repository question with a harness and without one

Guides

How AI Coding Agents Work: Model Plus Harness

Aug 19 · 13 min

Seven quantised coding models measured against one laptop's RAM and VRAM ceiling

Comparisons

Best Local LLM for Coding: Fit Decides It, Not Quality

Aug 18 · 12 min

A cache fee and a cache rent plotted against session length, crossing at 53 minutes

Comparisons

Gemini vs Claude Coding: The Price Flips at 53 Minutes

Aug 18 · 13 min

Nine repository files measured in tokens against the four-characters-per-token estimate

Guides

Context Window in AI Coding: What Actually Fills It

Aug 18 · 12 min

One measured coding turn priced on nine vendor cards, showing a 10.5x spread

Comparisons

Best AI Model for Coding: What the Price Cards Say

Aug 17 · 13 min

The Opus-to-Sonnet cost ratio decomposed into model, token shape and billing rate

Comparisons

Claude Sonnet vs Opus for Coding: The Ratio Is 2.5x

Aug 17 · 12 min

The SWE-bench three-stage filter applied to a real repository, yielding zero instances

Guides

SWE-bench Explained: What the Score Actually Measures

Aug 17 · 12 min

Muhammad Kashif — Founder & Editor — Page 7 · Devventa