---
title: Fewer Tokens per Task Is OpenAI's Pitch for GPT-6.2 as ChatGPT Moves to Token-Counted Limits
description: OpenAI says GPT-6.2 Astra and Sol do more within usage limits and cuts Astra's API price 20%, as ChatGPT and Codex move to one token allowance in December.
author: Darie Nani (Editor-in-Chief)
date: 2026-11-17T18:52:46.167Z
updated: 2026-11-17T19:01:04.255Z
canonical: https://www.sovereignmagazine.com/article/openai-gpt-6-2-astra-sol-usage-limits
image: https://cdn.nanimediahouse.com/pexels-a-software-developer-working-on-code-at-a-dual-monitor-setup-36706460.jpg
categories: Artificial Intelligence, Business
content_type: News
region: United States
publication: Sovereign Magazine
about:
  - type: Organization
    name: OpenAI
    description: American artificial intelligence company, developer of the ChatGPT chatbot and the GPT family of models.
---

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November and sold both models on how much work a subscriber can get out of a plan. The company says they "deliver better results on demanding tasks with fewer output tokens, so you can get more done within your usage limits."

The announcement does not mention it, but the way those limits are counted changes in December. From then, Plus, Pro and Business plans will share one allowance across ChatGPT and Codex, OpenAI's coding agent, and the allowance will be measured in tokens. "More advanced models and complex tasks generally use more of your allowance," OpenAI's [help page on the change](https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits) says. Quick replies in Instant chat will not count.

For developers who pay by the token, the same claim comes as a price cut. Astra, the frontier model, is priced in OpenAI's API at $8 for each million tokens sent to it and $40 for each million it writes, 20% less than GPT-6 Astra, and cached input is 60% cheaper at $0.40. Sol, the everyday model, keeps GPT-6.1 Sol's $2 and $10. Astra now costs the same per token as Anthropic's Claude Fable 5.2, which [Anthropic cut by 20% a week earlier](https://www.sovereignmagazine.com/article/claude-fable-5-2-price-free-resets).

## OpenAI Measures GPT-6.2 by What It Costs to Finish a Task

Most of the figures OpenAI published pair a score with a cost. On its own GDPval v2 test, built from multistep consulting, finance and design assignments, Astra scores 6.2 points higher than GPT-6 Astra while using 22% fewer output tokens. On Agents' Last Exam, a test of multistep professional work, OpenAI says Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points above Fable 5.2 at 5% of the cost. A footnote says Fable 5.2's leaderboard results leave out nine tasks that used a fallback model.

OpenAI says Sol "lets you complete more tasks on the same paid plan" than GPT-6.1 Sol, and that "many employees at OpenAI use it as their default model for both technical and non-technical tasks."

Artificial Analysis, which runs independent tests of AI models, ranks Astra second on its Intelligence Index, behind Fable 5.2 and ahead of Google's Gemini 4 Carbon and Anthropic's Claude Opus 5.5. It puts Sol "on par with GPT-6 Astra at ~15% of the Cost per Task". On its coding agent index, run through Codex, it found Astra level with GPT-6 Astra and Sol three points below GPT-6.1 Sol.

Both models are in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. OpenAI's release notes say the rollout "does not add either model to regular Chat."

## OpenAI Has Changed Its Codex Limits Several Times Since July

Since July, a bug report on OpenAI's Codex project on GitHub has collected complaints from Plus subscribers that "Codex usage accounting appears to drain the 5-hour limit abnormally fast". "One prompt just cost me 50%, clearly not fixed," one user wrote on 8 July. Comments kept arriving into October, and the report is still open.

OpenAI has changed the limits several times this year. In July it temporarily removed the five-hour limit for Plus, Business and Pro plans, and in August it brought the limit back for Plus. Tibo Sottiaux, who leads engineering for Codex and ChatGPT, wrote on X that the five-hour limit "allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage". On 29 September OpenAI announced a smaller Work and Codex allowance for its $200 Pro plan by email, and on 29 October it [canceled that reduction](https://www.sovereignmagazine.com/article/claude-ultra-plan-5-hour-limit) and gave the $100 and $500 Pro plans twice as much Work and Codex usage.

OpenAI is not alone. Google tightened Gemini's limits for AI Pro and AI Ultra subscribers in May, GitHub moved Copilot to metered credits in June, and Anthropic faces a [proposed class action over the limits on its Claude Max plans](https://www.sovereignmagazine.com/article/claude-max-lawsuit-ruling).

## From December, the Model a Subscriber Picks Decides How Fast the Allowance Goes

Under the shared allowance, a subscriber who spends a morning in Codex has less left for ChatGPT that afternoon, and the other way around. How much of the allowance a request uses "will depend on the model you choose and the task you ask it to do," OpenAI says. That makes the choice between Astra and Sol a choice about how long the allowance lasts. OpenAI also sells a faster Ultrafast mode, and when it introduced Ultrafast its release notes said it "uses your included allowance first, then your credit balance after that allowance is used." With GPT-6.2, Ultrafast reaches the $200 Pro plan for the first time, on Sol only, while the $500 plan gets it on both models.

> "We expect most users won't need to wait for their usage to reset more often than they do now."
> — OpenAI help page on the December change

Subscribers who reach the limit can wait for it to reset, upgrade their plan or buy credits. On OpenAI's developer forum, subscribers have started a thread titled "Please Keep ChatGPT and Codex Usage Limits Separate in December 2026".

## OpenAI Is Betting on Cost per Task Against Claude Fable 5.2

Artificial Analysis says OpenAI "now trails only Anthropic's Claude Fable 5.2" on its index. OpenAI's own charts set Astra beside Fable 5.2 mostly on cost per task, the measure on which its figures put Astra furthest ahead.

The launch comes as both companies raise money. Anthropic is expected to price its initial public offering on 18 November. OpenAI was reported at the end of October to have [lowered the target for its funding round to $20 billion](https://www.sovereignmagazine.com/article/is-the-ai-bubble-bursting-october-2026), and it aims to list in 2027.

How far GPT-6.2's lower token use stretches a Plus or Pro subscription will show once the shared allowance starts in December and subscribers can see their combined usage.

## FAQ

**Q: Is there a limit on how much you can use ChatGPT?**
Yes. Paid plans have an allowance for advanced models in ChatGPT Work and Codex, with a weekly limit and, on Plus, a five-hour limit. From December, Plus, Pro and Business plans share one allowance across ChatGPT and Codex, measured in tokens, so larger models and longer tasks use more of it. Instant chat does not count. At the limit, subscribers can wait for the reset, upgrade or buy credits.

**Q: What is the 5-hour usage limit on ChatGPT?**
It is a limit on how much Work and Codex usage a subscriber can spend in five hours, on top of the weekly limit. OpenAI temporarily removed it for Plus, Business and Pro plans in July and brought it back for Plus in August, saying it spreads the load on its computing capacity. OpenAI says the five-hour window does not apply to its Pro plans for now.

**Q: Does Codex offer unlimited usage?**
No. Codex usage comes out of the subscriber's plan allowance, and from December that allowance is shared with ChatGPT and counted in tokens, so the model and the size of the task decide how fast it goes. Of the Pro plans, the $500 plan includes the most usage, and Plus, Pro and Business subscribers can buy credits once the allowance is used.

Find out more at [openai.com](https://openai.com/index/introducing-gpt-6-2-astra-sol/).
