OpenAI's $100 ChatGPT Pro Has 3 Things Nobody Mentions. Anthropic Already Countered.

A hidden model. A usage limit that expires in 7 weeks. And a counter-move Anthropic shipped the same day. OpenAI's new $100/month ChatGPT Pro plan dropped on April 9, 2026, and the headlines all say the same thing: "5x more Codex usage." That's the boring part. Here are the three things every developer should actually care about.

The $100 Pro Tier: What You Actually Get

OpenAI finally filled the gap between their $20/month Plus plan and the $200/month Pro plan. The new $100 tier sits right in the middle, targeting developers who hit Codex usage limits but can't justify doubling their spend.

The numbers break down like this:

• ChatGPT Plus ($20/month) — Baseline Codex usage. Good for light coding sessions.
• ChatGPT Pro ($100/month) — 5x the Codex usage of Plus. Designed for "longer, high-effort Codex sessions."
• ChatGPT Pro ($200/month) — 20x the Codex usage of Plus. The power-user tier.

All three tiers include unlimited access to GPT-5.4 and GPT-5.4 Pro. The difference is purely about how much Codex you can run.

Codex now has over 3 million weekly users. That's a 5x increase in just three months, with 70% month-over-month growth. OpenAI isn't guessing that developers want more Codex capacity. They have the data.

Thing #1: GPT-5.3-Codex-Spark Is a Hidden Model You Can't Get Anywhere Else

Buried in the announcement is access to GPT-5.3-Codex-Spark, a research-preview model exclusive to $100+ Pro subscribers. You cannot access this model through the API. You cannot access it on Plus. It exists only inside ChatGPT Pro.

Codex-Spark is a smaller, faster variant of GPT-5.3-Codex. OpenAI built it for real-time coding. The key spec: over 1,000 tokens per second on ultra-low-latency hardware. That's not a typo. One thousand tokens per second. For comparison, standard GPT-5.4 runs closer to 100-200 tokens per second in most configurations.

It ships with a 128k context window (text-only at launch) and runs on separate infrastructure from standard Codex. Your Codex-Spark usage doesn't count against your normal Codex limits during the research preview.

Why does this matter? Because OpenAI is testing a model optimized for the specific latency profile that makes AI coding feel like pair programming instead of waiting for batch jobs. If Codex-Spark works as described, it changes how real-time code completion feels at the editor level.

The catch: it's in research preview. Usage limits may shift based on demand. And there's no timeline for API availability.

Thing #2: The 10x Usage Boost Expires May 31. The Real Limit Is 5x.

Every headline says "5x more Codex usage." But through May 31, 2026, $100 Pro subscribers actually get 10x the usage of Plus. That's double the advertised rate — a launch promotion designed to hook developers before the real limits kick in.

After May 31, the $100 plan drops back to 5x. If you signed up expecting 10x capacity as your baseline, you'll hit walls in June.

This is a textbook adoption strategy. OpenAI wants developers to build workflows around high Codex usage during the promotional window. By June, you're locked in. Dropping to 5x feels like losing something. Loss aversion does the rest.

The $200 tier doesn't have this problem — it's 20x from day one with no promotional asterisk.

If you're evaluating the $100 plan, test it now while the 10x promotion runs. But budget your workflow for the 5x reality that starts June 1.

Thing #3: Anthropic Shipped a Counter-Move the Same Week

OpenAI's $100 Pro plan is a direct response to Anthropic's $100/month Claude Pro, which includes Claude Code. OpenAI's blog post explicitly frames this as competitive positioning against Claude.

But Anthropic didn't sit still. The same week, they launched the Advisor Strategy — a fundamentally different approach to the pricing war.

The Advisor Strategy pairs Claude Opus 4.6 as a strategic advisor with Claude Sonnet 4.6 as the executor. Sonnet handles all tool calls and code generation at its lower price point. When Sonnet hits a decision above its pay grade, it consults Opus for guidance — typically just 400-700 tokens of strategic direction.

The results on SWE-bench Multilingual: Sonnet with Opus advisor scored 74.8%, up from 72.1% for Sonnet alone. That's a 2.7 percentage point improvement. The cost? 11.9% cheaper per agentic task than running Sonnet by itself.

Read that again. Better performance and lower cost.

The Advisor Strategy is available via the Claude Platform API with the advisor-tool-2026-03-01 beta header. You declare advisor_20260301 in your Messages API request, set max_uses to control cost, and the model handoff happens inside a single /v1/messages request. No extra round-trips.

The Pricing War Nobody Expected

Six months ago, the AI coding tool market had two tiers: free (GitHub Copilot for individuals) and expensive (enterprise contracts). Now you have four serious options at $100/month:

1. ChatGPT Pro $100 — 5x Codex, GPT-5.3-Codex-Spark access, Mac desktop app
2. Claude Pro $100 — Claude Code, Opus 4.6 access, Advisor Strategy for API users
3. Cursor Pro $20 — IDE-native, multi-model support, cheaper but different category
4. GitHub Copilot Enterprise $39/user — Team-oriented, org-level customization

The pricing is converging. OpenAI and Anthropic both landed on $100/month as the developer sweet spot. The question isn't price anymore. It's which model fits your coding style.

If you write Python and JavaScript and want fast iteration, Codex with Spark's 1,000 tokens/second latency is compelling. If you work on complex multi-file refactors and want an agent that reasons about architecture, Claude Code with the Advisor Strategy has the benchmark edge.

The Mac Desktop App Nobody Noticed

OpenAI also shipped a dedicated Codex Mac desktop app alongside the $100 announcement. It barely got coverage.

The app gives developers native macOS access to Codex without running through the browser. For developers already using the ChatGPT Mac app, this adds a dedicated coding workspace. For developers using VS Code extensions or Cursor, it's an alternative that ties directly into OpenAI's model ecosystem.

The strategic play here: OpenAI wants Codex on your desktop, not just in your IDE. If Codex lives in its own app, it becomes a standalone coding assistant rather than a feature of someone else's editor.

What This Means for Your Workflow

If you're currently on ChatGPT Plus ($20) and hitting Codex limits daily, the $100 Pro plan is an obvious upgrade. The 10x promotional window gives you until May 31 to test whether the expanded capacity justifies the 5x price increase.

If you're on Claude Pro ($100) and satisfied with Claude Code, there's no urgent reason to switch. Anthropic's Advisor Strategy gives you a path to better performance at lower cost through the API — something OpenAI hasn't matched yet.

If you're on the $200 OpenAI Pro plan, check your actual usage. The 20x limit is generous, but if you're consistently under 5x, you're paying $100/month for headroom you don't use.

The QVAC SDK offers yet another path: local AI coding that runs on your hardware with zero subscription cost. It won't match GPT-5.4 or Opus 4.6 on benchmarks, but for developers who want privacy and offline capability, it's worth watching.

The Bottom Line

OpenAI's $100 ChatGPT Pro plan is a well-executed pricing move. The 5x Codex capacity fills a real gap. GPT-5.3-Codex-Spark is genuinely interesting for real-time coding. The 10x promotional boost is clever customer acquisition.

But Anthropic's Advisor Strategy is the more technically innovative response. Better benchmarks at lower cost, with a clean API implementation that developers can integrate today. The pricing war is good for developers. The feature war is better.

Verdict: if you're choosing today, try the $100 Pro plan during the 10x window. Test Codex-Spark. Then compare to Claude Code's performance on your actual codebase. The right answer depends on your code, not the press release.

Frequently Asked Questions

What is GPT-5.3-Codex-Spark?

GPT-5.3-Codex-Spark is a research-preview model from OpenAI optimized for real-time coding at over 1,000 tokens per second. It's exclusive to ChatGPT Pro subscribers ($100/month and above) and is not available through the API. It uses a 128k context window and runs on dedicated low-latency hardware, separate from standard Codex infrastructure.

How does ChatGPT Pro $100 compare to Claude Pro $100?

Both cost $100/month and target developers. ChatGPT Pro gives 5x Codex usage (10x through May 31) plus exclusive Codex-Spark access. Claude Pro includes Claude Code and access to Opus 4.6. On benchmarks, Anthropic's Advisor Strategy (Sonnet + Opus) scores 74.8% on SWE-bench Multilingual at 11.9% lower cost per task. ChatGPT Pro offers faster token generation via Codex-Spark. The best choice depends on whether you prioritize speed or architectural reasoning.

Is the ChatGPT Pro $100 plan worth it in 2026?

If you're on ChatGPT Plus ($20) and hitting Codex limits daily, yes. The 5x usage increase (10x through May 31) removes the primary bottleneck. If you use Codex fewer than 5 sessions per week, the Plus plan is likely sufficient. Test during the promotional window before committing — the real 5x limit kicks in June 1.

What happened to the $200 ChatGPT Pro plan?

The $200 plan still exists and now offers 20x Codex usage compared to Plus. It includes all the same models and features as the $100 tier but with significantly higher usage limits. OpenAI repositioned it as the power-user option, with the new $100 tier filling the gap for developers who need more than Plus but less than the maximum.

What is Anthropic's Advisor Strategy?

The Advisor Strategy is an API feature where Claude Opus 4.6 acts as a strategic advisor to Claude Sonnet 4.6 (the executor). Sonnet handles all tool calls and code generation. When it hits a complex decision, it consults Opus for 400-700 tokens of guidance. This achieved 74.8% on SWE-bench Multilingual — 2.7 points above Sonnet alone — while costing 11.9% less per task. It's available via the Claude Platform API with the advisor-tool-2026-03-01 beta header.