OpenAI Rolls Out $500-a-Month Pro 500 — Speed Becomes a Product

A New Tier: $500 a Month, 25x the Plus Allowance

OpenAI unveiled Pro 500 at DevDay 2026 on September 29: $500 a month with a usage allowance 25 times the $20 Plus tier, making it the top individual subscription.

The centerpiece of Pro 500 is Ultrafast — OpenAI’s first-ever “speed tier,” built for GPT-6 Astra. In Codex, Ultrafast generates tokens up to 8x faster than the standard tier, officially rated at 300 tokens per second. On the API side, Ultrafast is a standalone service tier, up to 6x faster.

Note: that 8x refers to model output token speed — not to whole coding tasks getting 8x faster. Tool calls, test runs, and dependency waits are unaffected, so real-world gains will be smaller.

How Speed Is Priced

Ultrafast’s pricing logic is blunt: trade allowance for speed.

On subscriptions: using Ultrafast within Pro 500 burns subscription allowance at 8x the standard rate. On the API side: calling gpt-6-astra with service_tier: ultrafast is billed at 6x the standard rate.

There are also limits. Initial API quotas aren’t high: 500K tokens per minute for usage tiers 1–3, 1M for tier 4, 5M for tier 5. The official recommendation is to use WebSockets, otherwise per-request network overhead eats into the gains. Data residency supports the U.S. only — EU regional processing isn’t supported yet. An Ultrafast version of GPT-6.1 Sol is said to be “coming soon”; only Astra has one for now.

Also note: Ultrafast is currently available only in Pro 500 and Enterprise. Of the three new Pro tiers — Pro 100 ($100), Pro 200 ($200), Pro 500 ($500) — only the top tier gets Ultrafast.

Pro 200: Back on Sale, With Half the Allowance

Arriving alongside the new tier is another piece of news: Pro 200, which paused new subscriptions on September 10, is back on sale — with a shrunken allowance.

Included usage for ChatGPT Work and Codex dropped from 20x the Plus tier to 10x; weekly GPT-6 Pro messages fell from 200 to 100. The new allowance takes effect October 30. Eligible existing users can keep the old allowance until October 29, after which they’ll receive a one-time 62,500 usage credit (worth about $2,500), expiring at year end.

A quick bit of math: paying the same $200 gets you half the work after November. That’s the most controversial part of this adjustment.

What OpenAI Is Really Selling

In late 2024, OpenAI pushed the individual subscription ceiling to $200 for the first time; Anthropic and Perplexity quickly followed. Now the ceiling has been pushed to $500.

What deserves more attention is Ultrafast itself. Speed is being sold as a standalone SKU for the first time. The underlying logic is simple: latency matters most for interactive coding and agent loops — when a person is waiting on a reply, or a chain of tool calls is waiting on the previous step’s output, every second saved is money saved. Ultrafast’s speedup multiples (6–8x) roughly match its price multiples (6–8x): OpenAI isn’t selling cheaper compute, just more expensive time.

For you, the test is equally direct: if you live in Codex all day, or run long agent chains where every step waits on the model’s reply, Ultrafast saves time; if it’s everyday Q&A and writing, the standard tier is enough.

Sources: DevDay 2026 coverage from Aistify, Metir AI, and AI Copilot, plus OpenAI’s Pro tier documentation.