A generated hero card headed 'The fast lane got slower' with the subtitle 'Ultrafast reached ChatGPT Pro $200 for the first time — on GPT-6.2 Sol only', three cards reading 'Up to 4.5x speed on GPT-6.2 Sol and Astra', '6x the included usage', and 'Up to 8x on GPT-6 Astra, the model it replaces', and a footer reading 'Speed and usage multipliers per OpenAI documentation, 18 November 2026'.
Guides & Insights

GPT-6.2 Sol Ultrafast Reaches $200 Pro — and Burns 6× the Allowance for 4.5× the Speed

Author

jinhao.song

Date Published

Latest models · 20View all models →
Benchmarks: Artificial Analysis · updated daily
Back to all posts

GPT-6.2 Sol is the first model O​penAI has put behind Ultrafast on a $200 ChatGPT Pro plan, and the tier it unlocks is not the one the previous generation sold. O​penAI's own speed documentation says Ultrafast on GPT-6.2 Sol and GPT-6.2 Astra "generates tokens at up to 4.5 times the speed of Standard on the same model," while the same page still credits GPT-6 Astra — the model this one replaces — with "up to eight times its Standard speed." O​penAI's pricing page charges the new tier at 6× the included usage. Read those two sentences together and you get the shape of the whole launch: a wider door onto a slower lane, priced as if it were the faster one. Both models shipped on 17 November 2026, so this is a week-one read on a real product, not a rumour — but the 4.5× and the 8× are both vendor claims that nobody outside O​penAI has measured yet, and that gap is where the useful questions live.

The number the announcement did not put next to the other number

The launch material leads with access, and access is genuinely new. Ultrafast has existed since August 2026, but until this release it was gated to the most expensive plan O​penAI sells. Pro $200 subscribers were locked out. On 17 November that changed — partially, and the partiality is the story.

Here is what the ChatGPT-side documentation says, in its own words. Under Ultrafast mode: "On GPT-6.2 Astra and GPT-6.2 Sol, Ultrafast generates tokens at up to 4.5 times the speed of Standard on the same model." Further down the same page, describing the previous generation: "GPT-6 Astra and GPT-6.1 Sol also support Ultrafast on Pro $500 and eligible Enterprise and Edu plans. In Codex, GPT-6 Astra generates tokens at up to eight times its Standard speed with Ultrafast."

So the replacement tier is documented at 4.5×, and the tier it supersedes is documented at 8×. That is not a rounding difference or a marketing hedge — it is roughly half the speedup, stated on the same page, in adjacent sections, with no sentence connecting them.

Now the price side. The pricing page publishes speed multipliers for subscription usage, and they run the other way:

• GPT-6.2 Astra and GPT-6.2 Sol, Ultrafast — included usage consumed at 6×, and purchased credits or Enterprise pay-as-you-go at 3× through 28 February 2027, rising to 6× from 1 March 2027.

• GPT-6 Astra, Ultrafast — included usage at 8×, credits and pay-as-you-go at 6×.

• GPT-6.1 Sol, Ultrafast — the same 8× included and 6× credits.

• For comparison, Fast mode — 2× on GPT-6.2 Astra and Sol, and 2.5× included on GPT-6.1 Sol and GPT-6 Astra.

A screenshot of OpenAI's ChatGPT speed documentation page describing Ultrafast mode, showing that on GPT-6.2 Astra and GPT-6.2 Sol Ultrafast generates tokens at up to 4.5 times the speed of Standard, that Pro $500 subscribers get both models while Pro $200 subscribers get Sol only, and that GPT-6 Astra generates tokens at up to eight times its Standard speed with Ultrafast in Codex.

Four and a half over six is worse than eight over eight

Speed-per-unit-of-allowance is the only ratio that makes these two product generations comparable, and it takes one line of arithmetic.

• GPT-6.2 Sol Ultrafast — 4.5× the speed for 6× the allowance, so 0.75 units of speed per unit of allowance consumed.

• GPT-6 Astra Ultrafast — 8× the speed for 8× the allowance, so 1.0 units of speed per unit consumed.

• GPT-6.1 Sol Ultrafast — the same 1.0, on the same 8× multiplier.

On O​penAI's published figures, the new Ultrafast is 25% less efficient at converting subscription allowance into wall-clock than the tier it replaces. That is a defensible product decision — serving capacity is finite and a 4.5× tier is cheaper to keep online than an 8× one — but it is not what "faster than ever before" means, and a subscriber deciding whether to flip the switch deserves the ratio rather than the adjective. Both readings come from the same vendor, which is the only reason the comparison is available at all.

A generated single-column scoreboard for GPT-6.2 Sol Ultrafast listing documented speed of up to 4.5 times Standard, included usage consumed at 6 times, speed per unit of allowance at 0.75, promotional API rate of $6.00 input and $30.00 output per million tokens, list rate of $12.00 and $60.00, and US-only inference residency, with a footer noting that speed and usage multipliers are OpenAI's published figures and that no independent Ultrafast benchmark exists.

What Pro $200 unlocks, and the half of the line-up it does not

The access expansion is real but narrower than the framing suggests. From the GPT-6.2 release notes: "Pro $200 subscribers can use Ultrafast with Sol. Pro $500 subscribers can use it with either model, as can eligible Enterprise and Edu workspaces once an owner enables Ultrafast."

• Pro $200 — Ultrafast on GPT-6.2 Sol only. Astra, the stronger and more expensive model, stays on Standard and Fast.

• Pro $500 — Ultrafast on both GPT-6.2 Sol and GPT-6.2 Astra.

• Enterprise and Edu — both models, but Ultrafast is a separate permission that is off by default. A workspace owner has to enable it, per user or for the whole workspace, and per-user spend controls still apply.

• Plus, Business, and the consumer Chat surface — no Ultrafast. GPT-6.2 Astra and Sol are reachable in Work and Codex on those plans; they are not selectable in Chat at all.

There is a residency catch aimed at exactly the buyers who care most. Ultrafast on GPT-6.2 Astra, GPT-6.2 Sol and GPT-6 Astra supports inference residency in the United States only. GPT-6.1 Sol Ultrafast also covers Europe — the EEA and Switzerland. So the $200 subscriber who gains access this week is, if they need EU residency, losing ground relative to the model they were on.

The API side has a promo, and the promo is the actual deadline

Subscriptions get a fixed 6× multiplier. The API gets a discount with an expiry date, and the expiry is the most decision-relevant number in the release.

O​penAI's API rate card lists two Ultrafast rows per model. For GPT-6.2 Sol: $6.00 input, $0.30 cached input and $30.00 output per million tokens at the promotional rate, against $12.00, $0.60 and $60.00 at list. For GPT-6.2 Astra: $24.00, $1.20 and $120.00 promotional, against $48.00, $2.40 and $240.00 at list. Standard rates are $2.00 / $0.10 / $10.00 on Sol and $8.00 / $0.40 / $40.00 on Astra, which makes the promotional rows exactly 3× and the list rows exactly 6×.

O​penAI states the timing plainly: "Ultrafast costs 3 times the Standard rate through February 28, 2027, and 6 times the Standard rate starting March 1, 2027 (Pacific time)."

Work it through on a shape you might actually run — an agent session of 30,000 input and 1,500 output tokens per turn, over 40 turns, so 1.2 million input and 60,000 output, every request comfortably under the 272,000-token threshold where O​penAI reprices the entire request:

• GPT-6.2 Sol, Standard — $2.40 of input plus $0.60 of output. Three dollars a run.

• GPT-6.2 Sol, Ultrafast at the promo rate — $7.20 plus $1.80. Nine dollars a run, and roughly a third of the elapsed time if the 4.5× holds.

• GPT-6.2 Sol, Ultrafast at list — $14.40 plus $3.60. Eighteen dollars a run, for the same speed.

• GPT-6.2 Astra, Ultrafast at list — $57.60 plus $14.40. Seventy-two dollars a run.

That is the whole decision in one line: a pipeline switched to Ultrafast this month pays $9 a run, and the same pipeline in March pays $18. Anyone budgeting on today's number is budgeting on a price with fourteen weeks left on it. Above 272,000 input tokens it gets worse — input and cache rates double and output goes to 1.5× for the full request, which puts GPT-6.2 Sol Ultrafast at list at $24.00 input and $90.00 output per million.

A screenshot of OpenAI's API pricing page showing Ultrafast rates per million tokens for GPT-6.2 Sol at $6.00 input, $0.30 cached input and $30.00 output on the promotional rate and $12.00, $0.60 and $60.00 at list, and for GPT-6.2 Astra at $24.00, $1.20 and $120.00 promotional against $48.00, $2.40 and $240.00 at list, with a note that Ultrafast costs three times Standard through 28 February 2027 and six times from 1 March 2027.

Grading "worst release since the GPT-5 disaster"

The claim going around is a verdict, not a measurement, and it should be handled as one. What can be checked is narrower and more interesting than the jeer.

On Artificial Analysis — independent, and the only third-party measurement that exists for a model this new — GPT-6.2 Sol at max effort scores 57 on the Intelligence Index, ranked 11th of 91 models, at $1.01 per completed index task. GPT-6.1 Sol, which it replaces, scores 55 at $1.27 per task and sits 19th. So the successor is two index points and 20% cheaper per finished task, running at 85 output tokens per second against 49. On the headline independent numbers, this is a straightforward improvement, not a disaster.

Where the criticism has teeth is the interaction between the two Ultrafast decisions. A subscriber who reads "Ultrafast reaches Pro $200" and expects the 8× tier they have heard about gets 4.5×, on one of the two models, at 6× the allowance, with US-only residency. Every one of those four constraints is documented. None of them is in the sentence that announces the access. That is not a model failure — it is a launch-communication failure, and it is the kind of thing that produces exactly the aggrieved tone the post that flagged this release opened with.

Worth saying plainly: the model and the tier are separate products with separate numbers, and conflating them is how both get misjudged. GPT-6.2 Sol is a better model than GPT-6.1 Sol at the same rate card. GPT-6.2 Sol Ultrafast is a worse deal per unit of allowance than GPT-6 Astra Ultrafast. Both statements are true at once.

No independent measurement of Ultrafast exists, and that is the open question

Every speed figure in this piece is O​penAI's. There is no third-party Ultrafast benchmark for either generation, which means the 4.5× and the 8× have never been tested against each other under load, and the comparison above inherits that weakness.

The independent boards have not caught up either. Artificial Analysis's GPT-6.2 Sol page still renders a "No publicly available API" placeholder in its comparison chart while listing a live Intelligence Index, per-task cost and token speed in its summary panel — the placeholder is stale, not a claim that the model is unavailable. Anyone reading that chart as evidence of an unreleased model is reading it wrong, and anyone reading the 4.5× as measured rather than claimed is making the opposite mistake.

The honest position for the next few weeks: the model's quality is independently measured and improving, and the tier's speed is vendor-stated and unverified. Treat a 4.5× planning assumption as a ceiling until someone publishes a reproduction.

How to try it without betting an architecture on it

The reason this matters operationally is that Ultrafast is a service_tier flag, not a model swap. O​penAI documents it as service_tier: "ultrafast" on gpt-6.2-astra, gpt-6.2-sol, gpt-6-astra or gpt-6.1-sol in the Responses API, with separate rate limits from Standard and Fast mode, and a recommendation to use WebSockets so the latency gain is not eaten by connection overhead. That means the migration is a config change and the rollback is the same config change — which is the right shape for a tier whose price doubles in March.

OrcaRouter does not serve GPT-6.2 Sol or GPT-6.2 Astra today; the catalogue at api.orcarouter.ai carries openai/gpt-6-astra, openai/gpt-6-sol, openai/gpt-6.1-sol and openai/gpt-6-luna, and the GPT-6.2 model pages return 404. So this is not a recommendation to route the new model through us yet. What is true today is narrower and still useful: GPT-6.1 Sol is on the catalogue, it is the model whose Ultrafast tier already runs at 8× for 8×, and running your evaluation set against it is the cheapest way to find out whether a speed tier changes your product before you commit a budget to one that costs twice as much after February. Our pricing is the provider's list rate passed through with no markup, so when the promotional Ultrafast rate ends on 28 February 2027 the number on our side moves on the same day it moves on O​penAI's — no renegotiation, no repricing cycle, no surprise on a March invoice.

For the availability question, the accurate answer is that GPT-6.2 Sol and GPT-6.2 Astra are reachable through O​penAI's own API and third-party platforms generally, on the rate card above, with the plan gating described in this piece.

Who should turn it on this week

• Pro $200 subscribers running Codex or Work — the access is new and the promotional API rate does not apply to you; your included usage is consumed at a flat 6×. If you were waiting for the tier to become affordable, it became available, not cheap.

• Anyone with an interactive loop in it — 4.5× on a 7.9-minute index task is the difference between a coffee break and a context switch. That is the use case the tier is built for, and it is the one where the allowance math is least painful, because you are not running it all day.

• Anyone with a batch job and a morning — leave it on Standard. Batch runs at 50% of Standard, and Ultrafast at 6× pays six times as much to finish sooner than you needed it.

• Anyone needing EU residency — check before you switch. GPT-6.1 Sol Ultrafast covers the EEA and Switzerland; the GPT-6.2 models do not.

• Anyone budgeting past February 2027 — the API price doubles on 1 March 2027, Pacific time. Put that date in the ticket now, while it is still a footnote rather than a variance.

The launch is one honest sentence wrapped in three misleading ones. Ultrafast is now available to more people than it was last week, and the specific tier they got is slower and more expensive per unit of allowance than the one it replaces. Both halves are worth knowing before the switch gets flipped.

Compared in this article1

Detected from this article · Benchmarks: Artificial Analysis · updated daily