FRIDAY, NOVEMBER 20, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  AI
AI

OpenAI GPT-6.2 Astra and Sol Are Here, and the Price Cuts Are the Real Story

OpenAI released GPT-6.2 Astra and Sol on November 17. Astra costs 20 percent less than GPT-6 Astra, and Sol is the cheap everyday model at $2 per million input tokens.

OpenAI GPT-6.2 Astra and Sol Are Here, and the Price Cuts Are the Real Story
Source: FoxTPNL, Wikimedia Commons (CC BY 4.0)

OpenAI shipped two models at once

OpenAI released GPT-6.2 Astra and Sol on November 17, and the headline is not the benchmark chart. It is the price list. Astra, the flagship of the pair, costs $8 per million input tokens and $40 per million output tokens, which is 20 percent less than GPT-6 Astra. Sol, the smaller everyday model, costs $2 in and $10 out, the same as GPT-6.1 Sol. Both models are live in ChatGPT for Plus, Pro, and Enterprise users, and developers can call them through the API.

The timing matters. November has been a crowded month for model releases. Anthropic shipped Claude Fable 5.2 on November 9 at the same $8 and $40 price point, Google made Gemini 4 Hydrogen generally available on November 12, and Alibaba followed with Qwen4-Max on November 16. OpenAI is not launching into an empty room. It is launching into a price war, and Astra is its answer.

GPT-6.2 Astra pricing, in plain numbers

Astra is positioned as the model for serious coding and security work. OpenAI says coding and security performance improved over GPT-6 Astra, though the evidence so far comes from the company's own evaluations and early tester feedback, not independent benchmarks. Treat the performance claims the way you would treat any vendor benchmark: interesting, not settled.

The pricing is the part you can verify today. At $8 and $40 per million tokens, Astra now matches Claude Fable 5.2 token for token, which removes price as a reason to pick one over the other and puts the choice back on quality, speed, and tooling. Sol, at $2 and $10, undercuts most flagship-class models by a wide margin and is clearly aimed at high-volume work where a slightly weaker model is fine: drafting, summarising, triage, and the kind of coding tasks that do not need the biggest brain in the room.

Early reports also mention much larger cost reductions on some workloads, with one tracker citing cuts of 20 to 600 percent against earlier versions depending on the task. That range is so wide it tells you more about how different the workloads are than about the model itself. The concrete numbers are the list prices, and those are the ones to plan around.

Who should switch, and who should wait

If you are already building on GPT-6 Astra, the switch to GPT-6.2 Astra is close to free money. Same vendor, same API shape, 20 percent cheaper, and OpenAI claims better coding and security performance. Run your own evals on a slice of real traffic before you move everything, but there is no good reason to stay on the older model once you have confirmed parity on your workloads.

If you are choosing between Astra and Claude Fable 5.2, the decision got harder in a good way. Identical pricing means the tiebreakers are now the things that actually matter day to day: which model writes better code in your stack, which one handles your longest contexts without losing the thread, and which tooling you already have. Spend an afternoon running both on your real tasks instead of reading benchmark tables.

Sol deserves a look from anyone paying flagship prices for routine work. A lot of production AI spend goes to tasks that never needed a frontier model in the first place. If your pipeline sends every support ticket, every document summary, and every code review through the most expensive model you have, Sol at $2 per million input tokens is a straightforward way to cut that bill. The risk is quality drift on edge cases, so route the easy 80 percent to Sol and keep the hard 20 percent on Astra.

The bigger pattern

Step back and the pattern of November is clear. Every major lab is now competing on price per token as openly as on capability. Google priced Gemini 4 Hydrogen at $1.50 and $7.50 with a free tier. Alibaba priced Qwen4-Max at $1.50 and $4.50. Anthropic cut Fable 5.2 to $8 and $40. Now OpenAI has matched that price and added a $2 budget tier. For developers, this is the best market in the history of AI APIs, and it is still getting cheaper.

The catch is that cheaper models are only cheaper if you use them well. A model that costs a fifth as much but needs three retries to get the right answer is not a bargain. The teams saving real money this month are the ones measuring cost per completed task, not cost per token. GPT-6.2 gives them better numbers on both sides of that equation, but only if they actually measure.

OpenAI has not said anything about what comes after 6.2, and it does not need to. The message of this release is that the frontier is now a commodity market, and the labs know it. The winners will be the teams that pick the right model for each job instead of defaulting to the biggest one.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.