Key Takeaways
- Eleven days in, the Copilot pricing change looks worse, not better: GitHub's own documentation now lists GPT-5.5 at a 57x multiplier for legacy annual subscribers, and every follow-up prompt counts again.
- The community reaction is not mild confusion: the public GitHub discussion reads like a trust collapse, with developers questioning both the value and the way the change was handled.
- I was forced to test alternatives properly: Google Antigravity and OpenAI Codex have both proved more capable, more enjoyable, and much better value for my real work.
- This changed my AI stack: Copilot went from my default tool to the vendor that pushed me into better options.
It has now been eleven days since GitHub Copilot's new pricing model came into effect on June 1, 2026. My view has hardened, not softened: this was not just a clumsy billing change. It was a trust event.
In my previous article on Copilot's bait-and-switch pricing, I wrote about GitHub estimating that my April usage would jump from $39 USD to $1,941.07 under the new model. That sounded absurd before the switch. After using the new system, and after watching the public reaction unfold, it now feels like the estimate was the warning label.
The Billing Pain Is in the Repetition
The part that stings is not just that premium models are expensive. It is that the chargeable unit does not reflect how real agentic work happens.
GitHub's own legacy annual-plan multiplier table now lists GPT-5.5 at 57x. Claude Opus 4.8 is 27x. Gemini 3.5 Flash is 14x. These are not obscure edge models; they are exactly the sort of models developers reach for when the task is complex enough to justify an AI coding agent.
The bigger problem is how requests are counted. GitHub's request documentation says each time you send a prompt in a chat window or trigger a response, you are making a request. A follow-up prompt counts again. A steering prompt counts again.
If you ask the model to adjust the approach after seeing the first result, that is another full hit.
That matters because agentic coding is not a single-shot workflow. I do not hand an AI a prompt, walk away, and accept whatever comes back. I steer it. I correct assumptions.
I ask it to inspect a different file. I ask for a safer implementation plan. I ask it to rerun a failing test. That is the work.
Under this model, normal professional direction becomes an accumulating penalty. There is no practical allowance for the fact that much of the context is already cached in the conversation. There is no feeling that the product is sharing efficiency gains with the user. It feels like every natural step in the workflow has become a fresh toll.
The Forum Chatter Is a Trust Collapse
The public reaction has been profoundly negative. The official GitHub Community announcement and FAQ is not full of developers calmly asking for clarification. It is full of people questioning the value, the timing, the communication, and the motivation behind the change.
The strongest theme is scepticism. Developers feel baited. They feel Copilot encouraged heavy agent-mode workflows, normalised the use of high-end models, then changed the economics after habits had formed. Whether that was the explicit strategy or not, that is how it lands when the bill changes this sharply.
I still had about five weeks left on my annual subscription when the new policy hit. That is a poor customer experience in itself. Annual subscribers made a commitment under one value proposition, then watched the product economics change before the subscription period finished. GitHub can explain the billing mechanics all it likes, but the customer experience is simple: I paid for a year and got a materially worse deal before the year ended.
What makes this worse is that Copilot was not a weak product. I used it heavily because it was genuinely useful. The agent mode in VS Code was polished, fast, and productive.
That is why the change feels so cynical. The dependency was real because the product worked.
Antigravity 2.0 Is Better Than I Expected
The first product that forced me to rethink my stack was Google Antigravity 2.0. The timing is hard to ignore: Google announced a major Antigravity push at I/O on May 19, 2026, less than two weeks before Copilot's June 1 billing shift. I am not claiming inside knowledge. I am saying the timing looks very convenient.
Convenient or not, the product is outstanding. Google describes Antigravity as an agent-first development platform, and its current pricing page lists access to Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet and Opus 4.6, and gpt-oss-120b even on the individual plan.
The browser integration is the immediate practical win. Regression testing is faster because the agent can inspect the result in context, not just modify files and hope. That lowers the number of expensive reasoning turns needed to verify frontend work. For website design and WordPress development, where visual regressions are a constant risk, that matters more than leaderboard performance.
The planning model also feels better. Antigravity's goal-oriented workflow and implementation planning make the old IDE-integrated Copilot experience feel clunky. Copilot often felt like a clever assistant living inside the editor. Antigravity feels more like a workbench for directing agents.
The killer feature, though, is speed. Gemini 3.5 Flash has surprised me more than any model release in the last six months. Historically, "Flash" meant the cheap cut-down option. This version does not feel like that.
In my own work it has gone head to head with Sonnet 4.6 and GPT-5.4 on complex implementation tasks, and it often gets the result right first time.
It is currently enjoying a speed boost from Google that may not last forever, but even if it slows down, the direction is clear. Fast models are no longer just for cheap summaries and autocomplete. They are becoming serious coding models.
The Value Gap Is Absurd
I already had a Google AI Pro subscription for other related services and had not even realised it gave me another serious AI agent tool to work with. In Australia, that subscription costs me AUD$32.99 per month.
On my current usage, I can get about two solid days of work done in Antigravity before I hit the weekly limit. That is not unlimited, and I would prefer clearer quota visibility, but it is a usable professional backup and sometimes a primary tool. Compare that with the Copilot estimate from my previous article: $1,941.07 USD for the month if April's pattern repeated.
Extrapolating roughly, that AUD$32.99 Gemini investment would cost around USD$776 inside Copilot at the same relative usage rate. That is not a small pricing difference. That is a different commercial universe.
For a solo operator, this is not theoretical accounting. Tool cost comes directly out of project margin, and unpredictable metering makes fixed-price work harder to quote responsibly.
This is the problem Copilot now has. Once developers are forced to explore alternatives, they do not compare Copilot against a vague memory of worse tools from 2024. They compare it against the best current tools in June 2026. And the current alternatives are not weak.
Codex Is the One I Should Have Tested Earlier
The second product that changed my view is Codex. I cannot quite believe I had not looked at it properly before. OpenAI introduced the Codex app in February 2026, and it is far more than a coding chatbot.
OpenAI's current Codex pricing documentation lists a wide set of surfaces and features: Codex web, the local app, CLI, IDE extension, SDK and scripting workflows, automations, in-app browser, Chrome browser control, SSH remote connections, custom instructions, skills, plugins, app connectors, MCP, subagents, GitHub delegation, and more. That is not a small feature list. It is a platform.
A recent client revision job showed me what that means in practice. A client sent three separate emails with website changes, including image attachments and Word documents. The work required new page creation and site-wide revisions on a production WordPress site. My prompt was simple: "please read my recent emails from Greg and implement the revisions".
Because I had connected my Google services through Codex's plugin capability, Codex could read Gmail, extract the attachments, convert the Word document, reconcile the instructions against the current website, and apply the revisions. That is not a minor convenience. That is a completely different class of tool.
This is where Copilot now feels narrow. It is excellent inside an editor. Codex can operate across the working environment: repository, browser, Gmail, documents, local files, WordPress, deployment tooling, and scheduled tasks. For my custom application development and client website work, that broader operating surface is far more valuable than a polished chat panel inside VS Code.
I have since subscribed to the lower ChatGPT Pro plan, which costs me AUD$155 per month. So far, it looks like enough for a full month of serious work without hitting limits. If I do have an unusually heavy month, Gemini gives me a capable backup. That combined cost is still nowhere near the Copilot estimate that made me start looking elsewhere.
Copilot Forced Me to Open My Eyes
If this sounds like gushing, it is. Thank you, Copilot, for forcing me to pay attention.
That is the real strategic mistake GitHub made. The old pricing let Copilot stay the default because switching felt unnecessary. The new pricing forced a hard evaluation. Once I did that evaluation, I found two products that can do everything I was using Copilot for, often better, faster, and with more enjoyment.
Antigravity is fast, practical, and surprisingly capable. Codex is a treasure trove of features that I am still discovering. Both have changed how I think about AI-assisted development in ways Copilot no longer is.
Could Copilot still recover? Yes. It remains a strong product technically, and GitHub could rebuild trust with simpler pricing, clearer limits, fairer treatment of annual subscribers, and proper accounting for cached context. But trust is harder to restore than it is to destroy.
The lesson for me is simple: do not let one vendor become the whole AI workflow. The tools are changing too quickly, and the commercial incentives are too unstable.
For now, Copilot is no longer my default. It is the product that taught me to stop assuming the most familiar tool was the best one.
