---
title: 'In the News: November 1, 2026 (Morning)'
description: 'A 12-framework benchmark prices one agent-built app from $2.67 to $4.97 a run, and OpenAI will pool ChatGPT and Codex usage in December.'
canonical_url: 'https://darkfactory.dev/news/2026-11-01-morning'
markdown_url: 'https://darkfactory.dev/news/2026-11-01-morning.md'
collection: news
date_published: '2026-11-01T07:30:00-05:00'
date_modified: '2026-11-01T07:30:00-05:00'
---

# In the News: November 1, 2026 (Morning)


Choosing a framework now changes what an agent run costs, and one developer has put numbers on it. Separately, OpenAI says Codex usage will draw from the same allowance as ChatGPT chat starting in December.

## 1. Building the same app with Claude Code costs $2.67 to $4.97 depending on the framework

**[The AI cost of JavaScript frameworks](https://nolanlawson.com/2026/10/28/the-ai-cost-of-javascript-frameworks/)** · Nolan Lawson · Read the Tea Leaves, October 28, 2026

Lawson had a coding agent build one Pokémon app, using the PokéAPI, in twelve frameworks and meta-frameworks. He ran Opus 5.5 at high effort through Claude Code in a sandbox, five times per framework, and reports medians against a spec-conformance test suite the agent could see. Cost per run, priced at API rates, ran from $2.67 for Hono+htmx to $4.97 for Angular, with Astro ($2.76), Nuxt ($2.82), plain Vite ($2.98) and Next.js ($3.14) near the cheap end. Wall-clock time ran from 8.2 minutes for Astro to 17.2 for Angular. Lines of code ran from 1,370 for Hono+htmx to 2,009 for Angular, which he reads as the same app in roughly 70 percent of the code.

Two of Angular's five runs failed the spec because the agent built a server-rendered app that rejected non-localhost requests, despite instructions that it would sit behind a reverse proxy. Lawson did not re-run them, so Angular's figures come from three runs. He also records how rarely agents used the tooling each framework ships: TanStack Start's skill files were found in 2 of 5 runs, Next.js's skill file was read in 1 of 5 and its AGENTS.md in 2 of 5, and Nuxt's MCP server was never used. SvelteKit's higher cost may come from agents reading its v3 migration guide in every run, since it was too new for the model's training data.

The limits are his own. It is one content-heavy app, the results are stochastic, and the cost ranges overlap: React Router's cheapest run ($2.82) beat Hono+htmx's most expensive ($3.19). He also notes that Lighthouse performance and accessibility scores were nearly saturated across frameworks, so the differences are in cost, time, code size and JavaScript bytes. We read the post and its tables and did not rerun the benchmark.

**Why it matters:** If you pick a stack partly on agent cost, this gives you a measured spread of about 1.9 times between cheapest and costliest on one app, not a guess. The tooling numbers say that shipping a skill or MCP server does not mean the agent will find it, so put the instruction in the file the agent reliably reads.

## 2. OpenAI will share one usage allowance between ChatGPT and Codex from December

**[Understanding ChatGPT and Codex usage limits](https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits)** · OpenAI Help Center · page marked as updated the previous day

OpenAI's help page says that starting in December, the Plus, Pro and Business plans' usage allowance "will be shared across ChatGPT and Codex." Usage will be measured in tokens, and how much of the allowance a request uses depends on the model chosen and the task. Instant chat will not draw from it. The page says OpenAI expects most users "won't need to wait for their usage to reset more often than they do now," and that combined usage and reset times will appear under Settings, Usage. When the limit is reached, users can wait, upgrade or buy credits.

The page gives no allowance sizes, so whether coding-heavy users get less room cannot be told from it. A Hacker News thread on the page ([161 points and 97 comments when we looked](https://news.ycombinator.com/item?id=50320588)) is mostly commenters worrying about exactly that, including one who said not having chat and coding share a cap was a reason to stay with ChatGPT. Those are readers' reactions, not measurements.

**Why it matters:** If Codex is part of your build loop on a ChatGPT plan, a long agent run and non-Instant chat will soon draw on the same budget. Check the combined usage view once it appears, and keep a second provider or credits ready for the days an overnight run eats the allowance.