Summary
I estimated that 10 percentage points of the Codex Pro 500 weekly allowance represent about 1.426 billion tokens. Processing the same volume with GPT-6.1 Sol at Standard speed would use about 7,250 credits, equivalent to $202.99–$260.99 in purchased credits or $399.18 through the API. Allocating the monthly subscription fee across 30 days of available usage puts the subscription cost of the same volume at $11.67.
Using the allocated subscription cost as the baseline for the same volume of work, purchased credits cost 17.40–22.37 times as much and direct API usage costs 34.22 times as much.
Download the original Korean report (PDF)
Weekly usage and token counts
The analysis covers an account that switched to Pro 500 at 11:27 Korea time on November 6, 2026. At 06:58 Korea time on November 11, it had used 89% of the weekly allowance and had 11% remaining. The credit balance was the same at the start and end of the observation period.
I matched daily token counts from the OpenAI account usage view with weekly usage records close to midnight. I interpreted the dates as UTC because the results did not specify a time zone. On November 8 and 9, the account used 2,851,177,108 tokens, while weekly usage rose from 61% to 81%, a difference of 20 percentage points. Dividing the total token count by 20 and multiplying by 10 gives 1,425,588,554 tokens per 10 percentage points. OpenAI account usage documentation
| Date | Account tokens | Weekly usage | Tokens per 10 percentage points |
|---|---|---|---|
| November 8 | 1,839,869,295 | 61% → 73% | About 1.533 billion |
| November 9 | 1,011,307,813 | 73% → 81% | About 1.264 billion |
| Total | 2,851,177,108 | 61% → 81% | About 1.426 billion |

The November 6 total may include usage before the plan change. November 7 lacked a reading close enough to the start of the day. The November 10 daily token count was still being tallied when I checked it. For the November 8 and 9 date boundaries used in the analysis, the recorded times were approximately 21 seconds to 14 minutes from midnight.
The cost of processing the same volume with GPT-6.1 Sol
I held the estimated token count per 10 percentage points constant and priced that volume using GPT-6.1 Sol. I used the proportions of uncached input, cached input, and output from 475 Sol responses totaling 73,634,457 tokens recorded on this device on November 8–9. The direct API estimate also accounts for the rates on 73 Sol requests with more than 272,000 input tokens. Codex credit rates · OpenAI API pricing
| Processing speed | Credits | Cost at credit purchase prices | Direct API |
|---|---|---|---|
| Standard | 7,250 | $202.99~260.99 | $399.18 |
| Fast | 14,499 | $405.98~521.97 | $798.35 |
| Ultrafast | 43,498 | $1,217.94~1,565.92 | $2,395.06 |
Recorded usage by model and cost differences
In the responses recorded on this device, Sol accounted for 70.98% of the tokens on November 8–9 and 88.75% of all tokens collected after the switch to Pro 500. GPT-6 Astra accounted for the remaining 29.02% of the two-day sample. The earlier Standard API estimate using the mix of the two models recorded on those same two days was $962.23.
The daily totals from OpenAI’s Codex account view cover all tokens on the account. At 00:00 UTC on November 10, the usage screen showed 80.6% of the weekly allowance used, of which 61.8 percentage points were attributed to Sol and 17.6 percentage points to Astra. The screen warned that some usage might be missing. The figures by model show deductions from the weekly allowance. OpenAI’s API organization usage view reports input, cached input, and output tokens by model for API calls made by that organization. The share of tokens used by each model across the personal Codex subscription was unavailable. Codex account usage · API organization usage
| Sol share | Astra share | Standard API cost per 10 percentage points |
|---|---|---|
| 100% | 0% | $399.18 |
| 95% | 5% | $496.19 |
| 90% | 10% | $593.19 |
Depending on whether the server dates are interpreted as UTC or Korea time and whether Sol responses are drawn from the two-day sample or the entire period after the switch, the Standard API estimate for an all-Sol workload ranges from $268.48 to $399.18 per 10 percentage points. The figures exclude tool fees, taxes, and separate API credit benefits. API prompt caching documentation · API credits for subscribers
Allocating the monthly subscription fee to the same usage
Assuming the full weekly allowance is used, the equivalent allowance over 30 days is about 4.29 times the weekly allowance (30 ÷ 7). Ten percentage points of weekly usage are 2.33% of that 30-day volume, equivalent to $11.67 of the $500 monthly fee. The 89% weekly usage recorded at 06:58 Korea time on November 11 represents 20.77% of the 30-day volume, equivalent to $103.83 of the subscription fee. Codex subscription pricing
| Weekly usage | Allocated subscription cost | Sol cost at credit purchase prices | Sol direct API cost |
|---|---|---|---|
| 10%p | $11.67 | $202.99~260.99 | $399.18 |
| 20 percentage points over the two days analyzed | $23.33 | $405.98~521.97 | $798.35 |
| 89% at the observed time | $103.83 | $1,806.61~2,322.78 | $3,552.68 |
Conclusion
For the same volume at Sol Standard speed, purchased credits cost 17.40–22.37 times the allocated subscription amount, and direct API usage costs 34.22 times that amount. Direct API usage costs 1.53–1.97 times the equivalent price in purchased credits. For 10 percentage points of the weekly allowance, the direct API estimate is about $388 above the allocated subscription cost. At 89% of the allowance, the difference is about $3,449. The allocation is a unit cost that assumes full use of the 30-day equivalent allowance. The actual subscription charge is $500 per month. The replacement cost changes with the account-wide mix of models and the amount of cached input the API can reuse in practice.

“Codex Pro 500 Weekly Usage and Estimated GPT-6.1 Sol Costs”에 대한 2개의 댓글