---
canonical: https://tanveerriaz.me/blog/which-ai-model
format: text/markdown
---

I get this question constantly - in LinkedIn comments, at Singapore meetups, after Local AI sessions. "Which AI model is the best?" It's the wrong question. It's like asking who the best athlete is without naming a sport.

## Best at what job?

Frontier cloud models are brilliant at open-ended reasoning, long-context synthesis, and tasks where quality beats cost. Local models on Ollama are brilliant at privacy-sensitive drafting, offline work, and high-volume tasks where marginal cost matters. Coding agents excel at repo-aware edits. Small models excel at classification and routing. None of that fits on a single leaderboard.

## How I actually choose

For HalalEats RAG answers, retrieval quality and prompt structure matter more than whether the model is Claude or GPT. For Skillz agent procedures, consistency and tool compatibility beat benchmark scores. For SoundsHuman - local LinkedIn comment drafts - "good enough on-device with zero API cost" wins. For a board-facing summary, I'll pay for the frontier model every time.

Rule of thumb: List the job, constraints (privacy, cost, latency), and failure modes first. Then pick the model. Not the other way around.

## The 40–50% local insight

In our Local AI session we demoed that a free local model handles roughly 40–50% of everyday knowledge work - drafting, summarising, reformatting, document Q&A. Save paid frontier models for the hard half. That's not anti-cloud. It's portfolio thinking - same way you wouldn't run every banking workload on the same server tier.

## Stop chasing a single winner

Build a small roster: one local default, one frontier for hard tasks, one coding agent in your IDE, one fast model for routing. Match the model to the job. Update the roster when evidence says so - not when Twitter declares a new king.

Better question: "What's the job, and what's the harness?"

[Also on LinkedIn →](https://www.linkedin.com/feed/update/urn:li:activity:7466831673884717056/)