🔥 OpenAI Cut Its Flagship AI Price 20%, and Its Budget Model Nearly Caught It
OpenAI released GPT-6.2 Astra and Sol on 17 Nov 2026, cutting Astra's price 20% to $8/$40 per million tokens while Sol nearly matched the flagship at a fifth of the cost.

Everyone assumed the next frontier model would be a story about raw intelligence. OpenAI just told a different one, and it is written in dollars.
On 17 November 2026, OpenAI released GPT-6.2 Astra and Sol, and the number that actually matters is not a benchmark. It is a 20% price cut on its most capable model, Astra, which now costs $8 per million input tokens and $40 per million output tokens (OpenAI Developer Community). The cheaper sibling, Sol, holds at $2 in and $10 out, and on a test of real-world coding it lands within half a point of the flagship (casrai).
So the headline writes itself: OpenAI made its best coding model a fifth cheaper to run and shipped a budget model that nearly keeps up for a fraction of the bill.
The frontier race just stopped being about who is smartest. It is now about who is cheapest per useful task.
🧠 Why This Matters
For two years, the pitch from every AI lab was the same: bigger model, higher score, pay up. GPT-6.2 breaks the pattern because the jump in capability is modest and the drop in cost is steep.
OpenAI says Astra now does the same work for less than a third of GPT-6 Astra’s cost per task, partly because the model finishes jobs using fewer output tokens (9to5Mac). When you are paying by the token, a model that talks less and gets to the point faster is worth more than one that scores a fraction higher and rambles.
That matters to you if you build with these tools, because your bill is a function of tokens, not intelligence points. It matters to OpenAI’s rivals because the floor just dropped. And it matters to the whole “AI is too expensive to be profitable” argument, because this is OpenAI voluntarily shaving its own margins to keep developers from wandering off to cheaper models.
📊 Deep Dive
The two models split the work. Astra is the one OpenAI points you at for the hardest problems, including a claim that it is “the best model in the world for scientists” (casrai). Sol is the everyday workhorse for coding and writing, and the pricing is where it earns its keep.
Here is how the release stacks up:
- Astra pricing: $8 per million input tokens, $40 per million output, which is 20% below GPT-6 Astra (OpenAI).
- Sol pricing: $2 input, $10 output, unchanged from GPT-6.1 Sol, so the value gain comes entirely from a better model at the same price.
- FrontierCode 2.0 (Cognition’s test of whether code changes are good enough for maintainers to merge): Sol comes within 0.5 points of Astra at maximum reasoning, for roughly one-fifth of the cost (9to5Mac).
- Agents’ Last Exam: Astra at its highest setting scores 5.2 points above GPT-6 Astra’s best, at less than half the cost per task (casrai).
- Ultrafast mode: generates tokens at up to 4.5x the standard rate, available to $200/month Pro users with Sol and $500/month Pro users with both models (9to5Mac).
Read the FrontierCode line again. A model that costs $2 and $10 gets within rounding distance of one that costs $8 and $40, on the single test that measures whether real engineers would accept the code. For a lot of production work, the expensive model just became optional.
“Super efficient and built to perform where it counts, the real world.” — OpenAI’s description of the release (9to5Mac)
⚠️ The Catch
Cheaper and more capable has a bill attached, and it is not financial.
Both GPT-6.2 models are rated Critical for cybersecurity in OpenAI’s system card, the company’s highest risk tier (casrai). OpenAI’s own line is that Astra is “our most capable model ever for cyber” (OpenAI), which is a selling point and a warning in the same sentence.
On an internal test of writing working exploit code, GPT-6.2 Astra succeeded 53.5% of the time and Sol 50.5%, against 31.5% for GPT-6 Astra (casrai). Those figures come from runs with the safeguards and classifiers switched off, so they measure raw capability rather than what you get through the front door. OpenAI’s answer is tighter default restrictions and a vetting programme, Trusted Access for Cyber, that sits alongside a biology equivalent called GPT-Rosalind.
The uncomfortable part: the same efficiency that makes Sol a bargain for your startup makes capable offensive tooling cheaper for everyone else. When the price of competent code generation falls, it falls for all uses at once.
🎯 What Happens Next
Watch the API dashboards, not the demos. The real test of GPT-6.2 is whether developers quietly swap expensive flagship calls for Sol and pocket the difference. If they do, OpenAI’s token revenue per user could fall even as usage climbs, which is a strange place for a company reportedly chasing a trillion-dollar valuation.
Expect the rivals to answer on price within weeks, because a 20% cut from the market leader is a dare, not a footnote. And expect the cyber vetting programmes to get a lot of scrutiny, since “Critical” is not a label regulators ignore.
🧩 Bigger Picture
Zoom out and the trend is clean: the cost of frontier-grade intelligence is collapsing roughly as fast as the capability is climbing. GPT-6.2 is the clearest sign yet that the AI business is turning into an infrastructure business, where the winner is not whoever has the smartest model for a week but whoever can serve good-enough answers at the lowest cost per task, forever.
That is good news if you are building on top of this stuff. Your costs are heading in the right direction. It is harder news for anyone betting that model quality alone builds a moat, because Sol just proved that the gap between flagship and budget can shrink to half a point.
The smartest model used to be the story. Now the receipt is.
Sources
- GPT-6.2 Astra and Sol are now available — OpenAI Developer Community
- OpenAI upgrades ChatGPT and Codex with new GPT-6.2 Astra and Sol models — 9to5Mac
- OpenAI releases GPT-6.2 Astra and Sol with cheaper API pricing and tighter cyber safeguards — casrai
❓ Quick answers
How much does GPT-6.2 Astra cost?
GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens as of 17 November 2026, which OpenAI says is 20% below GPT-6 Astra (OpenAI Developer Community).
What is the difference between GPT-6.2 Astra and Sol?
Astra is OpenAI's most capable model for the hardest work and science; Sol is a cheaper everyday coding and writing model at $2/$10 per million tokens that came within 0.5 points of Astra on FrontierCode 2.0 at about a fifth of the cost (9to5Mac).
Are the GPT-6.2 models considered a security risk?
Yes. Both GPT-6.2 Astra and Sol are rated Critical for cybersecurity in OpenAI's system card, OpenAI's highest risk tier, and ship with tighter default safeguards and a Trusted Access for Cyber vetting programme (casrai).





