Pricing Watch | OpenAI Adds a 6x Ultrafast Speed Tier, Halves Pro 200's Usage Allowance

🌏 中文版

Summary of Changes

OpenAI made two pricing moves in opposite directions at the same DevDay 2026 event: on one hand, GPT-6.1 Sol matches flagship-level performance at a fifth of the cost (see the 09-30 model card); on the other, the flagship GPT-6 Astra got a new "pay for speed" tier, Ultrafast, priced at six times standard. At the same event, ChatGPT Pro 200's Codex/Work usage allowance was cut in half, and the new Pro 500 plan ($500/month) is the only tier that bundles this new speed tier. OpenAI is effectively turning "how fast" into a third dimension of pricing, alongside "how smart" and "how cheap" — and for teams running Codex heavily, this update shifts a chunk of what used to be included in Pro 200 onto a pricier new tier.

Before & After

Astra's Ultrafast tier claims up to 6x faster generation via the API and up to 8x in Codex (300 tokens/second); usage draws from the plan's included allowance first, then from the credit balance. OpenAI says GPT-6.1 Sol's Ultrafast tier is "coming soon" — for now it's Astra-only.

Cost Estimate

Scenario: A team running agentic coding tasks on GPT-6 Astra via the API, processing 5 million input tokens + 2 million output tokens per day (short context), moving all traffic from Standard to Ultrafast.

That 500% increase buys speed, not quality — OpenAI's announcement doesn't claim Ultrafast scores higher on any benchmark; it's the same model, just faster. The subscription math tells a similar story: Pro 200's old 20x Plus allowance is worth roughly $400 in Plus-equivalent terms (at $20/month); cut to 10x, it's worth about $200. Existing subscribers effectively lose half their monthly included usage, and OpenAI's one-time $2,500 credit — while it could cover roughly 12 months of that gap — is only valid for about 3 months before it expires.

Impact on Developers & Enterprises

Who Benefits Most

Ultrafast targets latency-sensitive, user-facing workloads where speed itself translates into business value — real-time code completion, or interactive agents that need to finish multiple tool-call rounds within a single conversation turn. For those cases, paying 5x more for 6-8x the speed is a reasonable trade. But most background, asynchronous agent work — code review, test generation, document cleanup — doesn't need real-time responses, and switching it to Ultrafast just means paying 500% more for nothing.

Competitive Landscape

OpenAI didn't invent "pay for speed" — Google's Gemini API already has a Priority tier (roughly 1.8x standard price for priority processing) — but OpenAI's new markup is notably steeper:

OpenAI now splits speed into three price bands (Standard/Fast/Ultrafast) instead of two, slicing "speed" more finely — and giving itself more room to charge for it. For anyone comparing rate cards across vendors, this means you can no longer just look at the input/output unit price; you need to check which speed tier that price actually belongs to.

Action Items

• If you're running background batch work on Astra: stay on Standard. Ultrafast offers you nothing — it's purely extra cost.
• If you have a real-time, latency-critical use case: benchmark Ultrafast on a small slice of traffic first to confirm it actually hits the claimed 6-8x speedup before moving your whole pipeline over. Don't pay the 500% premium for a theoretical ceiling you never actually hit.
• If you're an existing Pro 200 subscriber: plan your workload around your current allowance before 10/29, and use the $2,500 credit before it expires on 12/31 — don't let it go to waste.
• If you're considering upgrading to Pro 500: calculate your actual Ultrafast usage needs first. For most teams, paying for Ultrafast on the API as needed is cheaper than committing to a flat $500/month subscription.

Expiration Notice

⏰ Existing Pro 200 allowance preserved until: 2026-10-29. Starting 10/30, existing subscribers' Codex/ChatGPT Work usage cap drops from 20x Plus to the new, lower cap (10x Plus). ⏰ One-time $2,500 usage credit expires: 2026-12-31. Any unused credit from the downgrade is forfeited after this date — it does not roll over.

Takeaway

It's easy to focus on GPT-6.1 Sol matching flagship performance at a fifth of the cost and miss that the same event also opened a "pay for speed" tier on the flagship model, priced six times higher. That's OpenAI making "how fast" a third pricing dimension alongside "how smart" and "how cheap." For agent developers, that means rate-card comparisons now need an extra column — not just input/output unit price, but which speed tier that price applies to — or you risk mistaking a Fast or Ultrafast quote for a normal Standard-tier cost.

References

• OpenAI: DevDay 2026 Recap
• Pricing | OpenAI API
• About ChatGPT Pro tiers | OpenAI Help Center
• TheNextWeb: OpenAI halves Pro 200 usage and launches a $500 ChatGPT plan at DevDay
• VentureBeat: OpenAI's GPT-6.1 Sol offers Astra-like performance at 1/5th price. A new Ultrafast tier clocks at 300 tokens per second.