OpenAI just dropped two new AI models on the same day — and the cheaper one might be the real story.
On Tuesday, November 17, OpenAI released GPT-6.2 Astra and GPT-6.2 Sol — Astra the new flagship, Sol the everyday model. Short answer: for heavy research or complex professional work, pick Astra. For almost everything else, Sol gives you nearly the same brain for a fraction of the price. But there is a catch in the fine print — and it is the part OpenAI did not put in its headline.
What is GPT-6.2 Astra?
Astra takes over from GPT-6 Astra as OpenAI's top model, and the first surprise is the price: API rates are down 20 percent, to $8 per million input tokens in and $40 out, from $10 and $50.
On Agents' Last Exam, a benchmark of long, multi-step professional tasks, it scored 5.2 points higher than GPT-6 Astra while costing less than a third as much per task.
Independent testing is more careful: Artificial Analysis gave Astra 62 on its Intelligence Index — behind Anthropic's Claude Fable 5.2 at 65, just ahead of Google's Gemini 4 Carbon at 60. On coding, though, Astra was flat — no improvement over GPT-6 Astra.
What is GPT-6.2 Sol?
Sol is the model most people should probably use. It scored 57 on the same index — exactly what GPT-6 Astra, the flagship of two months ago, earned — at roughly 15 percent of the cost per task.
API price is unchanged: $2 per million tokens in, $10 out. On Cognition's FrontierCode test, both passed 48.2 percent of tasks — but Sol cost $1.29 per run against Astra's $5.97.
One blemish: Sol slipped 3 points on coding-agent work versus GPT-6.1 Sol. Worth knowing if you live in your code editor.
GPT-6.2 Astra vs Sol: what is the difference?
| GPT-6.2 Astra | GPT-6.2 Sol | |
|---|---|---|
| Role | Flagship — hard professional and scientific work | Everyday — writing, coding help, research |
| Artificial Analysis Intelligence Index | 62 | 57 |
| API price (per million tokens) | $8 in / $40 out | $2 in / $10 out |
| Best for | Research, science, complex tasks | Everyday work at ~15% of Astra's cost |
Astra buys the top few points of capability; Sol buys nearly the same capability for a quarter of the token price.
Which one should you use?
It comes down to what you actually do with AI.
Choose Astra if your work is long, high-stakes and hard to redo — research, deep analysis, multi-step projects. That is where its lead shows up, and where the extra cost is rounding error next to getting it right. It is the kind of knowledge work AI is steadily taking on — see how AI and jobs are changing across the US.
Choose Sol if you write, brainstorm, summarize, plan or build with AI every day. For routine work the quality gap is small and the savings are not — heavy users will feel it in the bill before they feel it in the answers. Developers, think in cost per task, not cost per token — Sol's $1.29 per FrontierCode run against Astra's $5.97 is hard to argue with. Just keep Sol's small coding slip in mind for heavy agent work.
One more thing before you switch: both models carry OpenAI's "Critical" rating for cybersecurity capability under its own Preparedness Framework, and they shipped with tighter guardrails on higher-risk security help. Power cuts both ways: the same week, OpenAI's GPT-6 Cyber made headlines for turning software flaws into working exploits — see the biggest AI headlines this Monday morning. And with AI fakes getting harder to spot, know how to spot a fake video before you trust what you see. Treat these models as professional tools, not toys.
The bottom line
GPT-6.2 is not really about a faster flagship — it is about the middle catching up. Sol now does what the flagship did in September, for about 15 percent of the cost. Astra wins when the task is genuinely hard and the stakes are high. For everything else — the emails, the drafts, the daily grind — Sol is the smarter default, and your wallet will notice before your work does.
Try Sol for a week on everyday tasks, keep Astra for the work that matters, and see which one earns its keep.

Comments
Post a Comment