In Devsplainers' October 2026 comparison, the decisive question is not which AI subscription has the most impressive model list. It is whether you mainly ask short questions, run long coding agents, make images or video, or rely on research tools. The video compares ChatGPT Plus, Claude Pro and Google AI Pro with metered API access through OpenRouter. Its arithmetic is useful, but the headline figures need the right labels before they become buying advice.
Start with free access for occasional chat. For an API-backed workflow with uneven volume, compare the cost of your actual prompts with a monthly plan. For frequent coding-agent work, test Claude Pro or another coding plan against your weekly limits, not a theoretical dollar value. For image or video work, buy the app whose included tools and credits you will actually use. Review the checkout price and terms in your country before committing.
Watch the Full Comparison
Credit: Devsplainers' analysis and examples come from the video and supplied transcript. Plan terms, the original SemiAnalysis methodology and the links below were checked separately on 19 October 2026.
A Subscription and an API Key Buy Different Things
A consumer plan is access to an application and its features, subject to usage limits. An API account is a separate, metered product that charges for model input and output, and sometimes other tools. A ChatGPT Plus payment does not become OpenAI API credit; Claude Pro and Claude API usage are separate too. This is why a $20 app subscription cannot be compared with $20 of API credit just by counting messages (00:31).
OpenRouter exposes many models through one account and, on its Standard tier, charges a 5.5% fee when credits are purchased, not on every request. Provider list prices still depend on the chosen model. The video uses it as the pay-as-you-go alternative; it does not reproduce every part of the first-party apps, such as their native memory, voice, projects, research interfaces or bundled media tools (03:58). Compare the entire workflow, including model availability, latency, data handling and features, not only token price.
Subscriptions are metered too. Anthropic says Claude Pro and Claude Code share usage across web, desktop, mobile and terminal, with rolling session and weekly limits; there is no fixed message count. ChatGPT's pricing page also describes plan-specific limits. Those conditions can change, and the same prompt can consume very different amounts of usage as the context grows.
The 33-Chats-a-Day Number Is an Example, Not a Threshold
The video estimates that a $20 subscription overtakes metered API spend at around 33 short chats a day (02:17). That implies roughly 990 exchanges in a 30-day month and about $0.020 per exchange to spend the entire $20. A longer response, reasoning mode, attached document, repeated context or an agent calling tools can change the bill sharply. An API gateway fee and taxes move the line again. The video's lower break-even for heavier thinking is a scenario, not a published universal rate.
To calculate your own line, save the token counts and model for a representative week. Multiply input and output tokens by that model's current rates on the OpenAI or Claude pricing page, then add any router fee and tool charges. Compare that monthly estimate with the subscription's local checkout price and the app features you would otherwise lose. A free plan may make the paid comparison moot if it already covers your workload.
What the $1,178 Claude Pro Figure Actually Means
The video's punchiest line describes a maximally used $20 Claude Pro plan as nearly $1,200 of “compute” (05:27). The underlying SemiAnalysis study measures subscription limits for specific models and token types, assumes an agentic workload mix, and converts the estimated maximum tokens to API-list-price equivalent value. The study explicitly says there is no single plan value without specifying the plan, model and workload. Its figures are not a measured infrastructure bill, a promise that every user receives $1,178 in usage, or a $1,178 credit balance.
That distinction changes the buying decision. Heavy Claude Code users can get much more API-equivalent work from a subscription than from $20 of API credit, if they can make use of its five-hour and weekly windows. A light user cannot cash out unused capacity. Anthropic notes that an ANTHROPIC_API_KEY configured for Claude Code can instead cause API billing, so check which account is paying before comparing totals. The report's comparison with ChatGPT Plus is a snapshot of agentic usage and list prices, not a verdict on every coding task or app feature.
Where the Money Goes by Task
| Workload | Practical starting point | What to check before paying |
|---|---|---|
| Occasional chat | Try ChatGPT, Claude and Gemini free tiers first. | Which model, message limits, uploads, privacy settings and app features are actually available in your region? |
| Frequent coding agents | Trial Claude Pro or a coding-specific plan; compare with metered API usage. | Do your real tasks fit the shared session and weekly limits? Does the editor, review workflow and model quality justify the plan? |
| Images | Compare ChatGPT's Free/Go/Plus image access with the output you need. | Quality, generations, editing controls, export and whether the local Go price is lower than Plus. |
| Video | Compare Google AI Pro's Flow credits with a free Flow trial. | How many clips your chosen model and settings consume, eligibility and your local subscription price. |
| Research | Test the free research tools before upgrading. | Source quality, citations, quotas, files and whether deep research is included in your plan. |
For coding, Devsplainers favors Claude Pro and names GitHub Copilot Pro as a lower-price alternative (07:19). GitHub currently lists Pro at $10/month in the US, with AI-credit allowances and paid overage options; this is not unlimited usage. For images, the video favors ChatGPT and says Go may be sufficient (07:43); OpenAI's live plan page is the safer source for current feature differences.
For video, Google's Flow help page lists 50 credits per day without a subscription and 1,000 monthly credits with AI Pro, subject to availability and product terms (08:11). In Portugal, Google currently shows AI Pro at €21.99/month, not the video's US-dollar shorthand. Flow credit cost per generation varies by feature; 1,000 credits is not a fixed number of videos. For research, the useful test is whether you can verify cited evidence, not how confidently a plan advertises “deep” search (08:47).
Free Tiers and Promotions Are Worth Checking, Not Assuming
Free ChatGPT, Claude and Gemini access has limits but is often enough for casual use (02:40). The video also mentions student promotions, phone bundles, regional offers, temporary limit increases and cancellation discounts. These are time- and location-specific offers, not a dependable buying plan. Check the provider's current eligibility page and renewal terms before signing up. In particular, a US student offer should not be presented as available to readers in Portugal without local confirmation.
Read the renewal date and price at checkout; set a reminder before the promotional period ends. Devsplainers recommends testing monthly rather than taking an annual discount before you know which tool earns a place in your workflow (12:08).
A One-Month Decision Test
- Name the job. Chat, coding, images, video and research should be evaluated separately.
- Run five representative tasks on the free tier. Record output quality, wait time, failure points and what you could not do.
- Measure a typical week. For API work, log tokens and tool charges; for subscriptions, note any limits hit and unused features.
- Trial one paid plan monthly. Compare the end-to-end result and total spend, including taxes, with the free or API route.
- Keep or cancel on evidence. Prefer the plan that saves useful time on repeated work. Re-check pricing when your workload or the provider's limits change.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | The $20 plan and the headline API-equivalent claim |
| 00:31 | Subscription versus API tokens |
| 01:19 | Subscription usage meters |
| 01:54 | What $20 of short API chats may buy |
| 02:17 | The illustrative 33-chats-a-day break-even |
| 02:40 | Leaderboard context and free plans |
| 03:00 | Free-plan features |
| 03:22 | Free-plan trade-offs |
| 03:47 | ChatGPT Go |
| 03:58 | What API access leaves out |
| 04:34 | Model and hosting differences |
| 04:52 | Coding-agent workload |
| 05:27 | SemiAnalysis subscription-value study |
| 05:56 | Heavy-user economics |
| 06:16 | Claude Pro limits and Fable |
| 06:40 | Usage meter versus API price list |
| 07:19 | Claude Pro and Copilot Pro for coding |
| 07:43 | Image-generation plans |
| 08:11 | Google AI Pro and Flow credits |
| 08:47 | Research tools |
| 09:16 | Student and device promotions |
| 09:43 | Regional giveaways |
| 10:00 | Temporary limits and models |
| 10:33 | Cancellation offers |
| 10:46 | Auto-renew risk |
| 11:29 | The video's recommendations |
| 12:08 | Why to test monthly first |
Watch Next: Four Related Comparisons
These are companion videos from the creator, not extra evidence for the exact prices above. They cover higher-priced plans, changing coding limits, model routing and local AI.
Sources and Price Limits
Video and methodology: Devsplainers' comparison, supplied transcript, and SemiAnalysis' original methodology. Official terms: ChatGPT plans, Claude plans, Google AI Portugal plans, OpenRouter billing, GitHub Copilot plans, Flow credits, and the OpenAI and Claude API price lists. Context: the LMArena text leaderboard is a moving benchmark, not a test of your own work. Prices, limits and offers were checked 19 October 2026; the video was published 14 October 2026. No account-specific quota or checkout price was independently tested for this article.