One-Line Take

The days of juggling three windows are over: I work in Claude Code and call Codex and Antigravity through their CLIs, and at $140 a month a new model no longer means agonizing over where to move (author's view)

I opened all three, then ended up with one window

  • I used to open Claude Code, Codex and Antigravity directly
  • These days I only open Claude Code, and calling the other two through their CLIs is simply easier (author's view)
  • The commands are in my October post on wiring the CLIs together
  • My own usage dashboard shows the current split

Usage dashboard capture showing 5-hour and weekly limits, today, 7-day and 30-day token counts and daily bars for Claude, Codex and Antigravity

Made by HSL — capture of my own usage dashboard, as of November 5, 2026. Screen labels translated to English

  • Over the last 30 days, Claude used 4.93B tokens, Antigravity 85.4M and Codex 41.5M
  • That makes Claude about 97% of the total
  • Those counts include cache, which inflates Claude, but it is still about 94% on output tokens alone
  • They only cover logs on this PC, so automated calls that keep no log and the ChatGPT app are missing
  • There is an obvious objection: Claude spends its own tokens writing instructions and reading results
  • So I can't promise this saves tokens, but it does cut the hassle of switching windows

$140 a month, and $100 of it is Claude Code

Bar dividing $140 a month into Claude Code Max at $100, Codex Plus at $20 and Antigravity AI Pro at $20

Made by HSL — official US prices: Claude, ChatGPT Plus and Codex, Google AI Pro. AI Pro's official $19.99 is shown as $20

Service Plan Monthly What I use it for
Claude Code Max $100 All my everyday work
Codex ChatGPT Plus $20 CLI research and drawings; ChatGPT search is on the same plan
Antigravity Google AI Pro $20 CLI research
  • Claude Max at $100 gives five times Pro's usage (Claude pricing)
  • Claude's weekly limit reads 1% in the capture because it had just reset; right before the reset it was at 66%
  • Codex's 9% weekly figure is also low because it comes half a day after its reset
  • Even so, Codex and Antigravity together are under 3% of my 30-day tokens
  • So the objection writes itself: why pay for tools you barely touch?
  • The $20 Plus plan also pays for ChatGPT search, so that part stings less
  • The rest pays off in the new-model section below, not in this month's usage

Claude Code for work, ChatGPT for looking things up

Flow where a task goes to ChatGPT if it is a search and to Claude Code otherwise, with Claude Code calling Codex and Antigravity through the CLI

Made by HSL — the author's workflow

  • Everyday work all happens in Claude Code
  • When I need to look something up, I use ChatGPT
  • Of what I have used, ChatGPT organizes answers best and is the most accurate (author's view)
  • That is my impression, not a measurement
  • Outside tests don't take my side
  • In a February 2026 multiple-choice news test, GPT-5 scored 85.0%, Claude 4.5 Sonnet 90.4% and Gemini 3 Flash 95.6% (arXiv)
  • In the 2025 EBU and BBC study, 36% of ChatGPT answers had a significant issue, against 76% for Gemini and 30% for Perplexity (EBU report)
  • Neither tested today's models, and both looked only at news; I found no test of accuracy on developer lookups or of how well answers are organized
  • That does not prove my ranking wrong, but nothing proves it right either
  • Tidy is not the same as correct, so I open the link behind any number that matters

Waiting for Carbon to reach Antigravity

  • Google announced Gemini 4 Carbon on October 28 (Google)
  • As I noted in my launch post, it is starting with a subset of cyber defenders
  • Developers and consumers get "as soon as possible", with no date and no mention of Antigravity
  • When I checked on November 5, the Antigravity model table still listed only Argon from Gemini 4
  • I'm hoping Carbon gets added to Antigravity (author's view)

Timeline: Argon limited on September 30, in Antigravity on October 13; Carbon limited on October 28, not in Antigravity on November 5

Made by HSL — Argon announcement, Antigravity blog, Carbon announcement, model table

  • Argon is the precedent: it reached Antigravity 13 days after its limited release
  • Carbon was on day 8 on November 5, though nothing guarantees the same gap
  • The plan tier matters too: on October 13 the Antigravity blog said Pro and Ultra, while Google's developer blog named only Ultra
  • One user reported Pro being blocked at first and working a few days later (Hacker News)
  • So even after it arrives, when it opens on my AI Pro plan is a separate question

With all three, a new model is no dilemma

  • Because I use all three platforms, I don't have to agonize over where to move when a new model launches (author's view)
  • If it lands on a plan I already have, I just change the model name and call it
  • My Codex CLI model list already includes GPT-6.1 Terra from October
  • No moving subscriptions, and no setting everything up from scratch

While a crowd rushes toward a NEW MODEL billboard, one developer sits with a mug and watches a parcel drop into one of three mailboxes

New model day

  • A subscription is not a ticket to everything, though, and Carbon is exactly that case
  • Getting a model does not mean getting much of it: OpenAI's top model Astra is listed at roughly 5–45 local messages per five hours on Plus (Codex pricing)
  • So the accurate version is "I can try whatever arrives without any fuss", not "I get everything on day one"
  • That is what the $40 a month buys, and it is charged in quiet months too

This post was researched by CLI and reviewed by three AIs

  • When do I call Codex and Antigravity? This post is the example
  • Codex and Antigravity (on Gemini 3.8 Flash) did the research, and Codex's image generation made the mailbox picture above
  • A Google post Antigravity found led me to fix "Ultra only on October 13" in my draft
  • Codex found the 2026 test above, and I checked its numbers against the original
  • The finished draft went to Astra (Codex CLI), Fable (Claude CLI) and Argon (Antigravity CLI) unchanged
  • All three replied within 90 seconds, and Claude Code pulled their opinions together
Note From What I did
Doesn't say when Codex and Antigravity get called All three Added this section and the pricing table entries
Doesn't say which plan ChatGPT is on Fable, Argon Added to the pricing table
Says things "changed" with no before-and-after Astra, Fable Changed to "the current split"
November 10 estimate for Carbon is weak Astra, Fable Removed the date
"Saves no tokens" is too absolute Astra Changed to "can't promise"
Remove the "(author's view)" tags Argon Declined; they separate fact from opinion
  • Astra and Fable read the draft without web access and said they could not check the October announcements
  • So I use AI opinions to find what to fix, and check facts against the originals separately

So What's Next?

  • The first thing to watch is whether a Carbon row appears in the Antigravity model table
  • If its AI Pro column is checked, what I'm hoping for has happened
  • If only Ultra stays checked for a long time, my "no dilemma with all three" claim gets one more caveat
  • Google has not given a date

Q&A (Field Notes)

  • Q. Does calling them through a CLI add API charges?
    • For Codex and Antigravity, logging in with your subscription takes usage out of the plan's limits. Setup is in the CLI post
  • Q. How does that dashboard work?
    • It adds up token counts from each CLI's logs on this PC and reads limits from each service every 10 minutes. No prompts or file paths are uploaded
  • Q. When does Carbon reach Antigravity?
    • Unknown. Google's announcement does not mention Antigravity, and Argon's 13 days are a single precedent
  • Q. Why look things up in ChatGPT instead of Claude Code?
    • It feels the most organized and accurate to me. I have not measured that, so I still check the links behind the numbers