GPT-5.6 Sol Output Falls to $20 per Million Tokens—for Three Months

On August 21, 2026, OpenAI cut GPT-5.6 Sol’s standard short-context price from $5 to $4 per million input tokens and from $30 to $20 per million output tokens. According to Reuters’ report on the change, the reduction covers the OpenAI API and is rolling out to eligible credit plans in ChatGPT Work and Codex, while Pro, Plus and Business subscription prices remain unchanged.

The output rate therefore fell by one-third, while the input rate declined by 20%. An August 21 update from OpenAI describes the lower API and credit pricing as a three-month promotion, so the headline saving applies to metered usage rather than ordinary ChatGPT subscription allowances.

The $4 and $20 rates have a short-context boundary

OpenAI’s GPT-5.6 Sol model page lists standard prices of $4 per million input tokens and $20 per million output tokens and says the promotional rates will be available at least through November 21, 2026. It also states that prompts exceeding 272,000 input tokens are charged at twice the input rate and 1.5 times the output rate for the entire request.

The advertised figures therefore should not be applied to unusually large-context requests without adjustment. The reduction changes the price of each metered token, not the number of tokens a request consumes; reasoning settings, tool use and response length can still change the total bill.

For a clean comparison, the relevant question is what an identical volume of input and output tokens would have cost before and after the promotion. Because the two rates fell by different percentages, there is no single saving percentage that describes every workload.

Three workloads show the difference in savings

These conditional examples isolate text-token costs under standard short-context pricing. They are arithmetic illustrations, not measured customer invoices, and exclude cached-input pricing, tool charges and other adjustments.

• Input-heavy retrieval: Eight million input tokens and one million output tokens previously cost $70—$40 for input and $30 for output. The same volume now costs $52, saving $18, or about 25.7%.
• Balanced analysis: Five million input tokens and five million output tokens fell from $175 to $120. The $55 saving equals about 31.4%.
• Output-heavy generation: Two million input tokens and eight million output tokens dropped from $250 to $168. The $82 saving is 32.8%, close to the full output-rate reduction because generated tokens dominate the bill.

An application producing short classifications from large prompts will save closer to the 20% input reduction. A coding or analysis workload that generates long responses can approach the 33.3% output reduction, although its actual invoice will still depend on the token mix and any additional charges.

The reduction reaches API, credit and Bedrock usage

Direct OpenAI API customers receive the revised model rates. The announced scope also includes eligible credit plans in ChatGPT Work and Codex, but the realized cost change for those users depends on the applicable credit schedule rather than a reduction in a monthly subscription fee.

A separate Amazon Bedrock notice, posted August 21, lists the same $4 input and $20 output rates, identifies the cuts as 20% and 33.3%, and says the promotion is available at least through November 21. This gives Bedrock customers a provider-specific confirmation of the change.

The notices do not establish identical terms for every reseller, marketplace or negotiated enterprise contract. Outside the OpenAI API, eligible OpenAI credits and Amazon Bedrock, customers must rely on their provider’s published rates and contract terms.

ChatGPT subscription allowances remain separate

The promotion does not reduce ChatGPT subscription fees or automatically increase the number of messages included with a plan. OpenAI’s current GPT-5.6 plan guidance says the model continues to use existing ChatGPT limits, which depend on the plan and, for managed workspaces, workspace settings.

A Plus or Pro subscriber therefore does not receive one-third more usage merely because API output tokens became one-third cheaper. Subscription access is controlled by plan-specific allowances and reset rules, whereas API and eligible credit use is metered under separate commercial terms.

The same boundary applies to Business workspaces. GPT-5.6 Sol may be available within the plan, but administrators can control model access and usage remains subject to workspace rules; the temporary token-rate reduction does not rewrite those allowances.

November 21 is the next pricing checkpoint

The wording “at least through November 21” makes that date a minimum commitment, not a guaranteed final day. OpenAI could extend the offer or retain the lower prices, but the published notices do not promise either outcome.

For now, the confirmed position is that eligible standard short-context usage receives the $4 input and $20 output rates during the three-month promotional window that began August 21. The next material development will be whether OpenAI extends the promotion, makes the rates permanent or publishes replacement pricing after November 21.

Also read:

• Google Ads Quality Score Is Not an Auction Input—Fix These Three Signals
• OpenAI Claps Back Hard at Anthropic with GPT-5.4-Cyber: Real Cyber Defense for the Actual World, Not Just the Elite
• Sam Altman Offers Every YC Startup $2 Million in OpenAI Tokens for Equity