# Did OpenAI halve GPT-6 Sol's Codex quota? The math does not prove it

> A Codex subscriber reports lower API-equivalent value after GPT-6 Sol's price cut. That figure alone cannot establish fewer tokens or less work per subscription.

By BIG CHANGE Editorial

Published: 2026-09-23T17:53:07.326Z
Updated: 2026-09-23T17:53:07.326Z
Canonical: https://bigchange.ai/blog/gpt-6-sol-codex-quota-cut-claim

![Six identical metal tokens sit in a brass tray on a vintage shop scale beside a blank price tag on its own metal stand.](https://bigchange.ai/api/media/file/gpt6-sol-quota-price-scale-hero-v1.png)
AI-generated conceptual illustration by BIG CHANGE.

A [Reddit post](https://www.reddit.com/r/codex/comments/1wo8fhw/okay_they_literally_cut_our_quota_by_half_gpt_6/) claims OpenAI cut the value of a Codex Pro subscription in half after GPT-6 Sol arrived. Its table estimates roughly $13 of API-equivalent usage for each percentage point of weekly quota, down from about $25 with GPT-5.6 Sol.

The author acknowledges the price cut and describes nearly unchanged usage. API-equivalent value can fall by half for an identical token allowance when rates halve. Establishing whether subscribers receive fewer tokens or completed tasks requires another measurement. The post's figures leave that question open.

## The big change

- **What changed:** Local Codex logs expose tokens. Third-party tools convert them to API dollars, while OpenAI's subscription meter uses a separate allowance.
- **Why it matters:** Subscription buyers need to compare completed work for their monthly fee. A falling API-dollar estimate can change the apparent bargain without showing whether the monthly fee buys less work.
- **What to watch:** Whether providers expose records linking model, speed, token mix and cache use to allowance debits. Such records would help teams compare subscription value across releases and investigate disputed depletion.

## The 50% result appears before any quota assumption

OpenAI's current Standard, short-context API rates are $4 per million input tokens and $20 per million output tokens for GPT-5.6 Sol. GPT-6 Sol costs $2 and $10 respectively. These are the [published API rates](https://developers.openai.com/api/docs/pricing); GPT-6 Sol's [model page](https://developers.openai.com/api/docs/models/gpt-6-sol) places long-context pricing above 272,000 input tokens.

For a fixed workload totaling one million uncached input tokens and one million output tokens, spread across requests that each stay inside the short-context band:

| Model | Input calculation | Output calculation | API-equivalent total |
| --- | --- | --- | --- |
| GPT-5.6 Sol | 1M × $4 | 1M × $20 | $24 |
| GPT-6 Sol | 1M × $2 | 1M × $10 | $12 |

At Standard speed, OpenAI's [Codex credit table](https://learn.chatgpt.com/docs/pricing) gives 600 versus 300 credits for the same uncached input/output totals. The page says credit prices alone do not determine included subscription usage.

Repricing alone can produce this pattern. Reproducing the account's result would require its token mix and settings, which the post does not provide as a matched before-and-after dataset.

## What ccusage measures

The [ccusage Codex documentation](https://github.com/ccusage/ccusage/blob/main/docs/guide/codex/index.md) labels Codex support experimental. The tool reads `token_count` events from local session files, groups them by the model in `turn_context`, then applies model prices from a pricing dataset. Its report is an API-equivalent estimate; the documentation expressly distinguishes it from a ChatGPT credit balance.

The cost formula separates ordinary input, cached input and output. Reasoning tokens are included in output billing. Recorded Standard or Fast service tiers can also change the estimate. A lower dollar total per percentage point therefore combines two different records: locally priced tokens and OpenAI's subscription-limit movement.

For more detail on why cache hits and context length change token cost, see BIG CHANGE's [GPT-6 prompt-caching guide](https://bigchange.ai/blog/gpt-6-prompt-caching-agent-context-costs). Our [GPT-6 and Claude price comparison](https://bigchange.ai/blog/gpt-6-luna-opus-5-5-price-release-race) places the API rate change in the wider model market. Neither article supplies evidence about subscription quotas.

## What OpenAI publishes about included usage

OpenAI estimates 300 to 3,000 local GPT-6 Sol messages per five hours on Pro 20x, compared with 200 to 2,000 for GPT-5.6 Sol. Weekly limits may apply; the page gives no weekly token cap. These variable estimates depend on model, context, reasoning, tools, retrieval and caching. They cannot establish the before-and-after allowance for the Reddit account.

The September 22 [Codex changelog](https://learn.chatgpt.com/docs/changelog) announces GPT-6 Sol and Luna with lower token prices. It does not announce a reduction in included subscription usage.

## The evidence that would settle the claim

Three measurements answer different questions:

| Question | Minimum useful record |
| --- | --- |
| Did API-equivalent value fall? | The same token counts multiplied by each model's published price. The answer is already yes. |
| Does GPT-6 Sol consume the meter differently? | The same account and plan, matched tasks, fixed reasoning and speed, recorded input/cached/output tokens, and before/after dashboard percentages across repeated runs. |
| Was the plan quota itself cut? | The same model, settings and workload measured before and after the alleged change, across comparable reset windows. |

Tool calls, retrieval, context growth and cache state must be captured because they change usage. Repetition matters because two agent runs can take different paths through the same task.

The reported drop in API-equivalent value is consistent with repricing. Whether the subscription delivers less work remains unresolved. A matched record connecting tokens, task outcomes and meter movement would address that question; the calculation alone cannot confirm a quota cut or exclude a metering problem.

## Sources

- [Okay, they literally cut our quota by half. GPT 6 Sol got hit too.](https://www.reddit.com/r/codex/comments/1wo8fhw/okay_they_literally_cut_our_quota_by_half_gpt_6/) — Community report behind the claim. Its API-equivalent dollar table and account observations are evidence of what the author reported, not proof of OpenAI's subscription-meter design or a plan-wide change.
- [OpenAI: Codex pricing](https://learn.chatgpt.com/docs/pricing) — Current plan-message estimates, weekly-limit caveat, variability factors and Standard token credit rates. The page states that credit prices alone do not determine included subscription usage.
- [OpenAI: ChatGPT and Codex changelog](https://learn.chatgpt.com/docs/changelog) — Announces the GPT-6 Sol and Luna rollout in Codex and ChatGPT Work at lower token prices. It does not announce a subscription-quota reduction.
- [OpenAI API pricing](https://developers.openai.com/api/docs/pricing) — Current Standard short-context rates used for the worked calculation: GPT-5.6 Sol at $4 input and $20 output, and GPT-6 Sol at $2 input and $10 output per million tokens.
- [OpenAI: GPT-6 Sol model](https://developers.openai.com/api/docs/models/gpt-6-sol) — Corroborates GPT-6 Sol's model-specific rates, 272,000-token long-context threshold and Standard, Batch/Flex and Fast price distinctions.
- [ccusage: Codex data source](https://github.com/ccusage/ccusage/blob/main/docs/guide/codex/index.md) — Primary project documentation for experimental Codex log support, token deltas, model attribution, pricing inputs and service-tier handling. It explicitly calls the result an API-equivalent estimate rather than a ChatGPT credit balance.
- [ccusage: cost modes](https://github.com/ccusage/ccusage/blob/main/docs/guide/cost-modes.md) — Explains supplied costs and token-based calculation using Claude Code examples. The separate Codex guide supplies the Codex-specific mechanism; neither exposes subscription billing.