---
title: "T3 Code Usage Dashboard: What the $197K API Estimate Really Means"
date: "2026-11-08"
video_id: "7705605418009054477"
tiktok_url: "https://www.tiktok.com/@of1ai/video/7705605418009054477"
creator: "@of1ai"
category: "Coding"
tags: ["t3 code", "claude code", "codex", "ai coding costs", "opencode"]
excerpt: "A T3 Code dashboard shows $197,470.62 of AI coding in 30 days. It is a list-price API estimate, not a bill. What T3 Code is and what the clip does not show."
---
T3 Code is a free, MIT-licensed app that puts AI coding agents such as Claude Code, Codex, Cursor, Grok and OpenCode in one interface, using your own subscriptions. A TikTok from @of1ai shows its usage dashboard reading $197,470.62 for 30 days of coding. That number is labelled an "API estimate," meaning what the tokens would cost at list API prices, and no source shows what the team actually paid.

## What is T3 Code?

The t3.codes page describes T3 Code as free and MIT licensed, with a bring-your-own-subscription model. It does not resell tokens. You connect the agent accounts you already pay for.

The site lists Claude Code, Codex CLI, Antigravity, OpenCode, Cursor and Grok CLI as supported providers. The GitHub repository, under the `pingdotgg` organization, lists Codex, Claude Code, Cursor, Grok Build, OpenCode and Google Antigravity. When fetched it showed the MIT license, about 53.9k stars and 10.2k forks.

The video ends with "Thank you, Theo & the T3 Code team," "Super awesome system. We love using it," and links to t3.codes and github.com/pingdotgg/t3code. The description tags @t3dotgg. The video discloses no paid relationship, and none was found.

## Did they really spend $197K on AI coding in 30 days?

No. The figure is what their usage would have cost at API prices. The dashboard shows a total of $197,470.62 with "15,225 sessions · API estimate" and a tag reading "30-day total, API estimate." The description says 30 days "would have cost us $197K at API prices," which matches the label.

The headline caption, "$197K of AI coding in 30 days," drops that qualifier. Someone reading only that line could take it as money spent.

Under T3 Code's bring-your-own-subscription model, the team's real outlay would be their Claude and Codex subscription fees. The video does not say what those were.

## How the $197K breaks down

The dashboard lists four providers:

- Codex: $103,402.81 (52.4% of cost, 125B tokens)
- Claude: $92,253.86 (3,853 sessions, 46.7%, 273B tokens)
- Cursor: $1,401.75 (4,121 sessions, 0.7%, 1.95B tokens)
- Grok Build: $412.20 (2,153 sessions, 0.2%, 1.27B tokens)

The rows add up to the total, and the percentages match. A second view breaks the cost down by speed tier: Standard $105,969.97, Fast $18,131.47 and Ultrafast $73,369.19, which sum to $197,470.63, a one-cent rounding gap. A fourth tier, Premium, shows $70,206.72.

That view also shows 2.63B uncached input tokens, 1.40B output tokens and cache savings of $2,045,203.05. A daily chart runs from Oct 9 to Nov 7, and its Oct 19 tooltip shows a total of $11,466.84, with Claude Code at $8,452.16 and Codex at $2,945.57.

The arithmetic shows the dashboard agrees with itself. It does not confirm that the token counts or prices behind it are correct, and no outside source for them was available.

The usage panel's range selector reads "Past 24h" in one shot while the headline says 30 days. That screenshot is dimmed and partly covered, so it is likely an artifact of the edit, but the video never shows the 30-day tab and the total together.

## Which agents does T3 Code support?

The description says T3 Code "puts Claude, Codex, Cursor, Grok and OpenCode in one app." Both the website and the repository list all five. The usage panel in the video is filtered to "4 providers" and has no OpenCode row, so the cost figures cover only four of them.

## Usage limits, cost tracking and settings

The video shows three other screens:

- **Usage limits:** a Limits tab for Claude Code showing Session 100% left, Weekly 24% left and "Weekly · Fable" 15% left.
- **API cost:** the 30-day cost chart and speed breakdown described above.
- **Settings:** a sidebar with Providers, Threads, Browser, Devices, Scheduled tasks, Projects, Source control, Worktrees, Environments and Sharing & access.

The description also promises "cost by model and project" and settings "synced across our machines." The cost view on screen breaks costs down by provider and speed tier, not by model or project, and no sync action appears. The t3.codes page and the repository, as summarized when fetched, do not mention usage tracking, cost reporting or sync. Those features may be in docs or a newer build, but the video is the only evidence for them here.

## The video stack credit

The closing card reads "Edited by Claude with HyperFrames and Sunra." An earlier card titled "Our AI video stack" lists Blender as the scene system and "HyperFrames or Remotion" as the motion layer. The video shows no editing session, prompt or render log, so these are the poster's own credits.

## Bottom line

The $197,470.62 figure is a list-price API estimate across Codex, Claude, Cursor and Grok Build, and the team's actual spend is not disclosed. T3 Code itself is free, MIT licensed and supports all five agents the description names. The usage limits and cost views appear in the video, while cost by project and cross-machine sync do not.

## Sources

- [t3.codes, T3 Code product page](https://t3.codes): provider list, free pricing, bring-your-own-subscription model and MIT license.
- [pingdotgg/t3code on GitHub](https://github.com/pingdotgg/t3code): repository, license, star and fork counts, and provider list.
- [TikTok video by @of1ai](https://www.tiktok.com/@of1ai/video/7705605418009054477): the source video.


## Video description and verification notes

**The file is a 140.000 s vertical clip, 1080x1920, MP4 (isom) container, with one HEVC Main video stream at 30 fps and a video bitrate of 188,773 bit/s (container 196,322 bit/s). It is 3,435,647 bytes. ffprobe reports no audio stream in the downloaded file, although metadata.json lists acodec aac. A missing audio stream in the download does not prove the original TikTok is silent. The post timestamp is epoch 1794101080, which is 2026-11-08 UTC (upload_date 20261108). The handle is @of1ai, the channel nickname is James, and the sound credit is "original sound" by James. The description is about 480 characters. At metadata capture (2026-11-09T21:01:31Z) the video had 533 views, 15 likes, 0 comments, 6 reposts and 10 saves. I read six frames: frame_001, 010, 025, 040, 055 and 070. I assumed 2 s spacing, so these sit at roughly 0 s, 18 s, 48 s, 78 s, 108 s and 138 s. I did not find a transcript file in the staged artifact (transcript.txt and transcript.json both failed to open), so I have no spoken words and cannot check transcript end time against audio duration. Everything below rests on frames and the description. Frame 001 has the caption "AT API PRICES" and "$197K of AI coding in 30 days". It shows a usage panel with "Usage / All environments / 4 providers / Past 24h", a boxed total of "$197,470.62", "15,225 sessions · API estimate", and a pink tag "30-day total, API estimate". Rows below it read Codex $103,402.81 (52.4% of cost, 125B tokens), Claude $92,253.86 (3,853 sessions, 46.7%, 273B tokens), Grok Build $412.20 (2,153 sessions, 0.2%, 1.27B tokens) and Cursor $1,401.75 (4,121 sessions, 0.7%, 1.95B tokens). The word-by-word caption at the bottom reads "grand." Frame 010 is titled "USAGE LIMITS / See what's left on every plan" and shows a Limits tab for Claude Code with Session 100% left, Weekly 24% left and "Weekly · Fable" 15% left, plus a strip of daily percentages. Frame 025 is titled "API COST / What it would cost at API prices". It shows a 30-day daily-cost chart (Oct 9 to Nov 7) with an Oct 19 tooltip (Codex $2,945.57, Claude Code $8,452.16, Grok Build $22.67, Cursor $46.43, total $11,466.84). It also shows Uncached input 2.63B, Output 1.40B, Cache savings $2,045,203.05, and Cost by speed: Standard $105,969.97, Fast $18,131.47, Ultrafast $73,369.19, Premium $70,206.72. Frame 040 is titled "SETTINGS / Everything is configurable" and shows a sidebar (Providers, Threads, Browser, Devices, Scheduled tasks, Projects, Source control, Worktrees, Environments, Sharing & access) next to a "What you can set" checklist with Environments and Sharing & access unchecked. Frame 055 is titled "UTLYZE RESOURCES / Our AI video stack" and lists "Scene system: Blender" and "Motion layer: HyperFrames or Remotion". Frame 070 reads "Thank you, Theo & the T3 Code team", "Super awesome system. We love using it.", "t3.codes", "github.com/pingdotgg/t3code" and "Edited by Claude with HyperFrames and Sunra". The description says the clip is a thank-you to T3 Code that puts Claude, Codex, Cursor, Grok and OpenCode in one app, with usage limits, cost by model and project, and settings synced across machines. No terminal, code editor, or live run of the agents appears in any frame I read. No frame I read shows a sync action between machines, or OpenCode in the cost panel.**

## The $197K figure is an on-screen estimate, and the video labels it as one

The headline number is $197,470.62, shown in frame 001 with the sub-label "API estimate" and the pink tag "30-day total, API estimate". The description says 30 days "would have cost us $197K at API prices". That wording matches the screen: the panel computes what token usage would cost at list API prices. It does not say the team paid that amount. I read the two lines separately and they are consistent with each other. The title and description use "would have cost", which is accurate to the label. A viewer who reads only "$197K of AI coding in 30 days" in the frame 001 headline could take it as money spent, and the caption does not repeat the qualifier in that line.

The row figures add up. Codex $103,402.81 plus Claude $92,253.86 plus Grok Build $412.20 plus Cursor $1,401.75 equals $197,470.62. The percentages also fit: $103,402.81 is 52.4% of the total and $92,253.86 is 46.7%. The speed breakdown in frame 025 (Standard $105,969.97, Fast $18,131.47, Ultrafast $73,369.19) sums to $197,470.63, a one-cent rounding difference. That internal arithmetic shows the dashboard is consistent with itself. It does not show the token counts or prices are right. I could not reach any source for the underlying prices or sessions, so the $197K total is Unverified beyond its own screen.

One detail needs a note. Frame 001 shows the range selector as "Past 24h" while the headline says 30 days, and frame 025 shows "30 days" selected. The frame 001 selector is a dimmed, partly covered screenshot, so I treat it as a staging artifact of the edit and not as evidence about the total. I cannot tell from the frames which range produced $197,470.62, and I did not see a frame where the 30-day tab and the total appear together.

## The product claims match T3 Code's public pages only in part

The description says T3 Code "puts Claude, Codex, Cursor, Grok and OpenCode in one app". The [t3.codes page](https://t3.codes) lists Claude Code, Codex CLI, Antigravity, OpenCode, Cursor and Grok CLI as supported providers. The [GitHub repository](https://github.com/pingdotgg/t3code) lists Codex, Claude Code, Cursor, Grok Build, OpenCode and Google Antigravity. So the five named providers are supported according to both pages. Frame 001 shows only four providers in the usage panel (Codex, Claude, Grok Build, Cursor) under the "4 providers" filter, and OpenCode does not appear in the cost rows. The description lists five. The usage panel shows four, so the cost figures cover four providers.

The description also names "usage limits, cost by model and project, every setting, synced across our machines". Frames 010, 025 and 040 show a Limits tab, a cost view with a speed breakdown, and a settings sidebar. The cost view I read breaks cost down by provider and by speed tier. I did not read a frame that shows cost by project or by model name, so "cost by model and project" is not shown in the frames I read. I also did not see any sync action. Neither the t3.codes page nor the repository page, as returned to me, mentions usage tracking, cost reporting, or settings sync. That does not show the features are absent. The fetch tool returns a summary of each page, and the features could be in docs or a newer build I did not read. I treat usage limits, cost views and sync as shown in the video only, and Unverified against first-party documentation.

## Open source, price and the "super awesome" thank-you

The t3.codes page describes the app as free and MIT licensed, with a bring-your-own-subscription model: it does not resell tokens. The repository page shows the MIT license, about 53.9k stars and 10.2k forks at the time I fetched it. That matters for reading the $197K figure. If users connect their own Claude and Codex subscriptions, the dashboard's "API estimate" compares subscription usage to list price. Under that model the team's actual outlay would be their subscription fees. The video does not state what they paid, and I found no source for it.

Frame 070 thanks "Theo & the T3 Code team", and the description tags @t3dotgg. The repository sits under the organization `pingdotgg`. I did not verify the connection between the poster and the T3 Code team, whether the poster has a sponsorship or affiliate arrangement, or whether the dashboard in frames 001 to 025 belongs to the poster's team or a demo. The description says "our machines" and "we love using it", which presents it as the poster's own usage. The video does not disclose a paid relationship, and I found none.

## The "Claude edited this whole video" claim

The description says "Bonus: Claude edited this whole video", and frame 070 says "Edited by Claude with HyperFrames and Sunra". Frame 055 shows a stack card headed "UTLYZE RESOURCES / Our AI video stack" with Blender as the scene system and "HyperFrames or Remotion" as the motion layer. These are the poster's own captions. I did not read any frame showing Claude doing the editing, such as a session, a prompt or a render log, and I did not fetch documentation for HyperFrames, Remotion or Sunra. The claim is Unverified, though the on-screen credit matches the description. Nothing in the frames shows the amount of human editing involved. Frame 055 also names "UTLYZE" as a brand, which I did not look up. Whisper mangling does not apply here because there is no transcript, so every product name above comes from burned-in captions.

## Key Takeaways

- **Clarification:** The $197,470.62 figure is labelled "API estimate" and "30-day total, API estimate" in frame 001. The description says it "would have cost" that much at API prices, so it is a list-price equivalent and not a bill.
- **Verified:** The four provider rows (Codex $103,402.81, Claude $92,253.86, Grok Build $412.20, Cursor $1,401.75) sum to the shown total, and the percentages of 52.4% and 46.7% match.
- **Partial correction:** The description names five providers including OpenCode. The usage panel in frame 001 shows "4 providers" and no OpenCode row. T3 Code's pages do list OpenCode as supported.
- **Unverified:** "Cost by model and project" and "synced across our machines" do not appear in the frames I read, and the t3.codes and GitHub pages as summarized do not mention usage tracking or sync.
- **Unverified:** "Claude edited this whole video" rests on the on-screen credit "Edited by Claude with HyperFrames and Sunra", with no editing session in the frames.
- **Context:** T3 Code is MIT licensed and free, with bring-your-own-subscription access. The video gives no data on what the team actually paid.
- **Context:** I had no transcript or audio stream. The analysis uses six frames (001, 010, 025, 040, 055, 070) and the description only.

## Resources

- [t3.codes, T3 Code product page, fetched 2026-11-09](https://t3.codes) establishes the provider list, free pricing, bring-your-own-subscription model, MIT license and absence of any named usage-tracking feature in the page summary I cite.
- [pingdotgg/t3code, GitHub, fetched 2026-11-09](https://github.com/pingdotgg/t3code) establishes the repository location, MIT license, star and fork counts, and provider list I cite.
- [TikTok video by @of1ai](https://www.tiktok.com/@of1ai/video/7705605418009054477) is the source video. I did not fetch the page. Title, description and counts come from the staged metadata.json.
