OpenAI now sells speed, and it costs $500

At DevDay on September 29, OpenAI did two things on the same afternoon. It launched Pro 500, a $500-a-month ChatGPT plan with the company's highest included usage and access to a new speed tier called Ultrafast. And it reopened Pro 200 to new subscribers with roughly half the usage the old $200 plan carried. Same price, less rope.

If you run anything on top of these models, that pairing is the whole story. The headline is the $500 number. The thing that will actually show up in your month is the quieter cut underneath it.

What the three tiers now look like

OpenAI's own help page lists three Pro plans: Pro 100 at $100, Pro 200 at $200, and Pro 500 at $500, with Ultrafast included only on the top one (1). Pro 500 is described as 25 times the ChatGPT Plus allowance, the highest OpenAI offers (6).

Pro 200 went from 20x Plus usage in Codex and Work down to 10x, and weekly GPT-6 Pro message caps in ChatGPT dropped from 200 to 100 (2). Existing subscribers are not hit immediately: if your Pro 200 was active at the eligibility cutoff or in the seven days before it, you keep your old allowance through October 29, 2026, then move to the smaller one at the same $200 price (1).

So there are really two questions on the table, and they are not the same question. One is whether $500 is worth it. The other is whether your current $200 still does the job after October 29.

Ultrafast is a throughput product, not a smarter model

Ultrafast does not make GPT-6 Astra better. It makes it faster. OpenAI describes it as a premium speed tier for workloads where speed matters most, generating up to about 300 tokens per second in Codex and Work, with the option also exposed to API users (2). In Codex that lands as roughly an 8x speed increase over standard mode (4).

Here is the part most coverage buried. Ultrafast draws down your included limits at 8x the standard rate, and purchased credits at 6x (6). Speed is not free capacity. You are trading allowance for latency. A 25x pool consumed at 8x is not a 25x pool anymore; it is closer to a 3x pool that answers immediately.

That framing matters because it kills the naive pitch. Pro 500 is not "more AI for more money." It is a latency purchase with a large usage buffer attached so the latency purchase does not strand you on day nine of the billing cycle.

Who should actually pay $500

Three profiles clear the bar.

People who are already capped. If you hit the old Pro 200 ceiling in Codex before the month ended, the cut to 10x is going to hurt on October 29 whether you upgrade or not. For you, $500 is not an upgrade, it is a repair.

People whose bottleneck is waiting. Agentic coding is a loop: read, propose, run, fail, revise. Latency compounds through every pass. If your day is twelve long Codex runs, cutting generation time changes how many complete loops fit into an afternoon. That is a real output number, not a vibe.

Teams billing the time out. One senior developer hour in the US costs more than the $300 delta. If the speed tier reliably returns two or three hours a month, the math ends there.

Who should stay at $200 or drop to $100

If your usage sits comfortably inside 10x Plus and nothing in your workflow is blocked on waiting, the extra $300 buys you nothing you will notice. All three Pro tiers now price roughly the same per unit of Plus-equivalent usage, so the jump buys scale and Ultrafast, not a discount (6).

That last point deserves emphasis because it changes how you should shop. There is no longer a volume reward for going up a tier. You are not getting cheaper tokens by spending more. You are getting permission to spend more, faster.

If you are a writer, a marketer, a solo founder doing three or four heavy sessions a week, you are probably a Pro 100 or Pro 200 customer who should ignore this launch entirely.

The rest of DevDay is not filler

A few of the other launches change the buying decision more than the pricing table does.

Every Pro subscriber gets one Dot, OpenAI's always-on personal agent, and conversations with your Dot do not count against your ChatGPT usage limit (2). That is an uncapped surface inside a capped product, which is worth understanding before you assume you need more allowance. Dots are not reaching Pro users in the EEA, Switzerland or the UK for now (5).

"Sign in with ChatGPT" lets Plus and Pro customers run OpenAI models inside 16 partner products, including Notion and Cognition, without setting up an API key (2). If your stack is three tools that each hold a separate key and a separate bill, your subscription just absorbed part of that.

ChatGPT Space replaces Library as a shared home for a team's files, built around pages that teammates, ChatGPT and dots can edit together, available on Pro, Business and Enterprise (5). OpenAI also shipped team workspaces, a meeting note-taker, @ChatGPT for Slack and Teams, and a cheaper Sol model.

How to decide in one week, not one month

1. Open your Codex and Work usage for the last 30 days and find the real number, not your impression of it. If you used more than 10x Plus, October 29 is your deadline, not a hypothetical.

2. Time three of your slowest recurring runs. Write the seconds down. That is the only honest input to the Ultrafast question.

3. Multiply the seconds saved by how often you run them, then by your billable hour. Compare it to $300.

4. Remember the 8x drawdown. If you plan to leave Ultrafast on all day, model your allowance at 8x consumption, not at 25x headroom.

5. Check what your subscription now replaces. Partner access through Sign in with ChatGPT may retire a paid seat somewhere else.

Mistakes to avoid

Assuming your $200 plan is unchanged. It is, until October 29, and then it is not. Grandfathering is a delay, not an exemption (1).

Buying credits on Pro 100 or Pro 200 to unlock Ultrafast. You cannot. Among Pro plans, Ultrafast ships only with Pro 500 (6).

Treating speed as quality. Ultrafast does not improve reasoning. If your outputs are wrong, a faster wrong answer is a worse product, not a better one.

Upgrading on feel. The only defensible reason to move to $500 is a number you measured: a cap you hit, or an hour you got back.

The read

OpenAI just split its power users into two groups and charged them differently. If your work is throughput-bound, $500 is a tool and the price is boring. If it is not, the reopened $200 tier with half the rope is the product they built for you, and the correct move is to notice the cut, check your usage, and stay exactly where you are.

The interesting shift is not the price. It is that speed is now a line item. Expect Anthropic and the rest to follow within a quarter, the way they followed the original $200 plan (2).