Dropping the smartest models for coding: tasks cost 20x less and limits stop mattering
ievkz · reddit · 2026-09-18
A programmer shares what happened after abandoning the strongest models. By "Cost per Task" (cost of tokens to complete a well-formed task), GPT-5.6 Luna (max) tops the chart at $0.18/task vs $3.26 for the smartest GPT-6 Astra (max) — nearly 20x — despite AAI intelligence indices of 53 vs 38. Switching to the cheapest Luna, he stopped hitting 5-hour and weekly limits on a $20/month plan, and Codex stopped going in circles. His conclusion: for skilled programmers, even the cheapest frontier models are smart enough; the intelligence race is nearing its end.
The next bottleneck is speed: mainstream models average 25-50 tokens/s on DeepInfra/OpenRouter, so agent tasks take 5-15 minutes; DeepSeek-V4.1-Flash via its own API hits 300 tokens/s, finishing tasks in under a minute with a slightly higher index than Luna. He argues the next agentic-AI race is speed, currently led by Chinese models, with OpenAI's future datacenters potentially pushing thousands of tokens/s — results ready in 10-30 seconds. He also likens this to the context-window arms race that already ended as meaningless past 1M tokens.
Related event: Dev ditches top model, finds nearly 20x cost gap per task(2 posts)→
More from coding & agent
- Jev Classifies 1.6k Bookmarks in 22s, 155x Faster and 10x Cheaper Than GLM 4.7 Flash — iannuttall · 2026-09-20
- Python Is Losing Steam Because AI Doesn't Need It to Be Simple — mark_k · 2026-09-20
- Dev rips apart Musk-boosted agent's harness: 'complete garbage' Python mess — MickeySteamboat · 2026-09-20
- Databricks Co-founder: If Starting a PhD Today, I'd Work on Reward Hacking — burny_tech · 2026-09-20
- Open-source tools track what AI coding agents really spend in tokens and cost — _jaydeepkarale · 2026-09-20
- AI Browser Agent Fills Out Immigration Card From Passport Photo and Email, Even Handles CAPTCHAs — TianbaoX · 2026-09-20