MeanCEO: Tech Startups and Startup Ideas

GPT-6.2 Astra Is 20% Cheaper and Now Costs the Same as Claude Fable 5.2: What OpenAI's New Models Cost, Where They Fall Short, and the EU Setting to Check

TL;DR: What did OpenAI release, and what does it cost?

On 17 November 2026 OpenAI released GPT-6.2 Astra, its new top model, and GPT-6.2 Sol, its everyday model. In the API, Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra and exactly the list price of Claude Fable 5.2. Sol stays at $2 and $10. Both run in ChatGPT and Codex for Plus, Pro, Enterprise and Edu users, in GitHub Copilot and, from day one, on OpenAI's EU data residency endpoint for a 10% surcharge. Independent tests show real gains in reasoning and science, but flat coding scores and high hallucination rates on hard questions, so test before you switch.
OpenAI just made its best model cheaper. It did not make it the cheapest.
GPT-6.2 Astra now has the same sticker price as Anthropic's Claude Fable 5.2, and GPT-6.2 Sol has the same price as Claude Sonnet 5.5. So the price war we wrote about in October has turned into something else: list prices are converging, and the real difference is how many tokens each model burns to finish your task. Here's what OpenAI announced, what the fine print says about long prompts, speed modes and EU hosting, what independent testers found, and how to decide whether to move your workflows.

What OpenAI released on 17 November

OpenAI's announcement of 17 November 2026 introduces two models. GPT-6.2 Astra is "our new frontier model for your hardest projects". GPT-6.2 Sol is the "efficient, versatile model for everyday work", and OpenAI says many of its own staff use it as their default. Both are available "to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex". The post doesn't mention the Business plan by name. In the API they're called gpt-6.2-astra and gpt-6.2-sol.
The headline is price. Astra's input and output tokens cost 20% less than GPT-6 Astra's, and cached input is 60% cheaper: $8 for input, $0.40 for cached input and $40 for output, per million tokens. Sol keeps GPT-6.1 Sol's rates of $2, $0.10 and $10. Both models have a context window of 1,050,000 tokens and can write up to 128,000 tokens in one answer, according to OpenAI's model pages. Astra's knowledge cutoff is 18 May 2026, Sol's is 1 June 2026.
On the same day, GitHub said both models are generally available in Copilot for Pro+, Max, Business and Enterprise users, "billed at provider list pricing under usage-based billing". If your team codes in Copilot, the new models are already in the model picker.
OpenAI's own benchmark claims are bold. It says Astra beats GPT-6 Astra on Agents' Last Exam, a test of multi-step professional work, "at less than a third of the cost" per task, and that Sol comes within 0.5 points of Astra on its FrontierCode 2.0 coding test at about one-fifth of the cost. These are OpenAI's numbers, measured by OpenAI. The independent ones come further down. Early testers were far less impressed: Matt Shumer called the models essentially unusable, and we collected the GPT-6.2 reviews from Shumer, Theo Browne, Ethan Mollick and Every.

The new price list, next to Claude and Gemini

Here's how the new models compare with their predecessors and their closest rivals, using each company's own list prices. We checked the OpenAI API pricing page, Anthropic's Claude pricing page and Google's Gemini API pricing page on 18 November 2026.
ModelInput per 1MCached input per 1MOutput per 1MNote
GPT-6 Astra (old)$10$1$50Replaced as OpenAI's top model
GPT-6.2 Astra$8$0.40$40New, 17 Nov 2026
Claude Fable 5.2$8$0.20$40Same list price as Astra, cheaper cache hits
GPT-6.2 Sol$2$0.10$10Same price as GPT-6.1 Sol
Claude Sonnet 5.5$2$0.10$10Same list price as Sol
Gemini 4 Carbon$2, then $4$0.10, then $0.20$10, then $20Price doubles on 1 Dec 2026
Standard short-context prices, US dollars. Claude's cached price is its cache-hit rate; OpenAI and Anthropic also charge for cache writes.
Two things stand out. First, at the top end, OpenAI and Anthropic now charge the same per token, so the cheaper model is the one that needs fewer tokens to finish your job. Second, Google's Carbon is the outlier for the next two weeks: as we explained in our post on Gemini 4 Hydrogen, Carbon's launch price doubles on 1 December.

The fine print: long prompts, speed modes and EU hosting

The per-token price is only the start. OpenAI's model pages and pricing page add several rules that can move your bill more than the 20% cut:
RuleWhat it does to your billApplies to
Long prompts over 272K input tokensInput and cache rates double, output costs 1.5 times as much, for the whole requestAstra and Sol
BatchHalf the Standard price, for jobs that can waitAstra and Sol
Background windows (beta)30% off for a 2-hour window, 40% off for an 8-hour window, on eligible requestsAstra and Sol
Fast modeTwice the Standard priceAstra and Sol
Ultrafast3 times Standard until 28 Feb 2027, then 6 times from 1 Mar 2027Astra and Sol, US hosting only
EU data residency10% more, on models released since 5 March 2026Astra and Sol via eu.api.openai.com
Sources: OpenAI API model pages, pricing page and data residency guide, checked 18 November 2026.
The EU part matters for anyone who promises customers that their data stays in Europe. OpenAI's data residency guide says "GPT-6.2 Astra and GPT-6.2 Sol support US and EU data residency with Standard, Fast, and Batch processing", through eu.api.openai.com, which covers the EEA and Switzerland. You need a key from a project created in the EU region, and OpenAI says data residency is open to "eligible organizations". Ultrafast works with US data residency only, so the fastest mode isn't an option if you've promised EU hosting. That's still an improvement: the same guide says Fast mode isn't available with EU data residency for GPT-6 Astra.
One more tip from OpenAI itself: to get the cheap cached rate, keep your prompt prefixes unchanged and append new instructions at the end. If your app rebuilds the system prompt on every call, you're paying full price for text the model has already seen.

What independent testers found

Independent benchmarks came out the same day, and they're more mixed than OpenAI's launch post. The Decoder summarised them on 17 November:
  • Overall: on the Artificial Analysis Intelligence Index, Astra scores 62, second behind Claude Fable 5.2 at 65. Epoch AI puts both at 174 on its capabilities index, a tie.
  • Cost per task: Vals AI measured Sol at $3.18 per task and Astra at $10.68, against $21.28 for GPT-6 Astra. Sol is the cheapest model in its top 10.
  • Coding: on Artificial Analysis's Coding Agent Index, Astra stays where GPT-6 Astra was, and Sol drops three points from GPT-6.1 Sol.
  • Hallucinations: at maximum effort, Astra guesses wrong instead of saying "I don't know" 59% of the time on Artificial Analysis's very hard factual test, and Sol 55%. The Decoder notes these aren't everyday rates.
OpenAI's own safety notes add two things worth knowing. The models launch with tighter safeguards on higher-risk cybersecurity help, which OpenAI admits "also restrict some legitimate vulnerability research and security testing workflows". And in two of OpenAI's tests, both models were "more likely to omit mentioning obstacles that blocked their work, such as a missing file or unavailable search tool". OpenAI says it added instructions in Codex that reduce this a lot. If you run them as agents through the API, ask for blockers to be reported explicitly.

What this means for you

  1. Don't switch on the headline. A 20% cut on Astra is real, but the same list price as Claude Fable 5.2 means the winner on your bill is whichever model finishes your tasks with fewer tokens. Run 20 of your own real tasks through both and compare cost per finished task. Our guide on how to evaluate an AI tool has a simple scoring sheet.
  2. Make Sol your default. At $2 and $10, Sol costs the same as GPT-6.1 Sol and Claude Sonnet 5.5, and independent tests put it close to Astra for a third of the cost per task. Keep Astra for long, high-value jobs.
  3. Watch the 272K line. Above 272,000 input tokens, the whole request gets more expensive. If you feed whole codebases or document piles into the 1M context window, split the work or trim the input. Our piece on why AI token costs rise 2 to 5 times shows where bills usually grow.
  4. Move waiting jobs to Batch. Overnight summaries, data cleaning and content updates can run at half price, or 30 to 40% off with background windows.
  5. Decide on EU hosting now. If you sell to European businesses, the 10% EU residency surcharge is often cheaper than losing a deal over a data transfer question. Set up an EU-region project before you build on the new models, because keys from a global project won't work there.
  6. Check your ChatGPT plan, not just the API. The new models are in ChatGPT Work and Codex, and from December your ChatGPT and Codex usage share one allowance. If Astra uses fewer tokens per task, you may get more done before you hit the limit.
My take: the most useful part of this launch isn't Astra's benchmark wins, it's that OpenAI, Anthropic and Google now sell their top models at nearly the same list prices. That's good news for bootstrappers, because it turns model choice into a test you can run yourself instead of a bet on whose marketing to believe. Earlier this autumn we covered the AI price war of October, GPT-6.1 Terra and Qwen4-Max at $1.50 per million tokens. The pattern holds: prices keep falling, models keep leapfrogging each other every few weeks, and the founders who win are the ones who can swap a model in an afternoon because they built a small test set and kept their prompts portable. For the rest of the month's AI releases, see our AI news for November 2026.
Are you getting customers from AI search yet?

More and more traffic is coming from Perplexity, ChatGPT, Grok and other AI tools.

👉 Rank on ChatGPT

FAQ: GPT-6.2 Astra and GPT-6.2 Sol

When were GPT-6.2 Astra and GPT-6.2 Sol released?

OpenAI released both models on 17 November 2026, in ChatGPT Work, Codex and the API. GitHub made them available in Copilot the same day.

How much does GPT-6.2 Astra cost in the API?

$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens at Standard rates. That's 20% less than GPT-6 Astra per token, and 60% less for cached input. Prompts over 272,000 input tokens cost more for the whole request.

How much does GPT-6.2 Sol cost?

$2 per million input tokens, $0.10 cached and $10 output, the same as GPT-6.1 Sol. Batch costs half of that.

Is GPT-6.2 Astra cheaper than Claude Fable 5.2?

Not per token. Both list at $8 input and $40 output per million tokens, and Claude's cache hits are cheaper at $0.20. OpenAI says Astra needs far fewer tokens per task, but that's its own measurement, so test on your own work.

Can I keep my data in the EU with GPT-6.2?

Yes, if your organisation is eligible. Both models support EU data residency through eu.api.openai.com with Standard, Fast and Batch processing, for a 10% surcharge. You need a key from an EU-region project, and Ultrafast is US-only.

Should I switch my startup's workflows to GPT-6.2?

Test first. Independent results show gains in reasoning and science, flat coding scores and high hallucination rates on hard factual questions. Run your own tasks through Sol, Astra and your current model, and compare cost per finished task.

About the Author

Violetta Bonenkamp, also known as MeanCEO, is an experienced startup founder with an impressive educational background including an MBA and four other higher education degrees. She has over 20 years of work experience across multiple countries, including 5 years as a solopreneur and serial entrepreneur. Throughout her startup experience she has applied for multiple startup grants at the EU level, in the Netherlands and Malta, and her startups received quite a few of those. She’s been living, studying and working in many countries around the globe and her extensive multicultural experience has influenced her immensely.
2026-11-18 04:17 News and Trends