Superpower Daily

Superpower Daily

Your reading space

Sign in
Huang Calls GPT-6 Astra AGI, but the Field Has No Shared TestCognition Releases SWE-2 for Devin, Claiming Lower-Cost Coding...Mariner Commits $175 Million to 700 AI Agents for Its Wealth-Ad...Pentagon Says It Has Moved 90% of Classified AI Work Off Anthro...Salesforce Adds AI Agents for Sales, Support and Internal Workf...Discovery Loop Reportedly Seeks $50 Billion Valuation in New Fu...Huang Calls GPT-6 Astra AGI, but the Field Has No Shared TestCognition Releases SWE-2 for Devin, Claiming Lower-Cost Coding...Mariner Commits $175 Million to 700 AI Agents for Its Wealth-Ad...Pentagon Says It Has Moved 90% of Classified AI Work Off Anthro...Salesforce Adds AI Agents for Sales, Support and Internal Workf...Discovery Loop Reportedly Seeks $50 Billion Valuation in New Fu...
Swipe

OpenAI Makes GPT-5.6 Sol Cheaper for Metered Use, Not Easier to Access

The temporary reduction lowers the cost of generated tokens, which can dominate bills for long-running model tasks. Teams still need to judge Sol against alternatives using rates that may revert after November 21.

By 2 min read
OpenAI Makes GPT-5.6 Sol Cheaper for Metered Use, Not Easier to Access
OpenAI Makes GPT-5.6 Sol Cheaper for Metered Use, Not Easier to Access

Listen to this story

The audio brief

About 1:29
0:001:29
Read transcript
OpenAI is cutting the metered price of GPT-5.6 Sol, with the biggest reduction landing on generated output—the part of long-running model tasks that can drive the largest bills. Through at least November 21, API output falls from 30 dollars to 20 dollars per million tokens. Input drops from 5 dollars to 4, and cached input from 50 cents to 40 cents. For a workload with one million input tokens and one million output tokens, that changes the bill from 35 dollars to 24, before tool calls, cache writes, priority processing, or higher rates for very large prompts. The economics are especially favorable for output-heavy workloads; tasks dominated by fresh input or cached context save less. But this is a pricing change, not an access expansion. Plus, Pro, and Business subscription allowances stay the same, as do five-hour and weekly limits. The reduction applies to API usage now, and is rolling out for eligible purchased-credit activity in ChatGPT Work and Codex. Those credits can extend qualifying paid activity, but they do not change what a subscription includes. Sol supports a 1.05-million-token context window and outputs of up to 128,000 tokens, so the discount could matter for unusually long jobs. Still, OpenAI can restore the old rates after November 21. The key question for any evaluation is whether Sol remains cost-effective once this promotional bill disappears.

Story brief

3 key points

OpenAI is temporarily cutting GPT-5.6 Sol’s metered rates through at least November 21, with output tokens dropping from $30 to $20 per million—the biggest savings for generation-heavy workloads. A balanced one-million-input/one-million-output task falls from $35 to $24 before extra fees. The lower prices apply to API usage and eligible purchased-credit activity in ChatGPT Work and Codex, but do not expand...

  1. 01

    Output pricing drops 33%, from $30 to $20 per million tokens; input falls 20% to $4, and cached input to $0.40.

  2. 02

    A balanced one-million-input/one-million-output workload costs $24 temporarily, versus $35 previously, excluding tools and other surcharges.

  3. 03

    Plus, Pro, and Business allowances, five-hour limits, weekly limits, and included ChatGPT usage remain unchanged.

OpenAI has lowered GPT-5.6 Sol’s metered pricing through at least November 21, with the sharpest cut landing on the generated output that can drive large model bills. The offer reduces API and eligible credit-paid use; it does not add capacity to ChatGPT subscriptions.

The promotion covers API calls and eligible activity paid with purchased credits in ChatGPT Work and Codex. Included usage, five-hour limits, weekly limits and subscription allowances for Plus, Pro and Business remain unchanged. For subscribers, purchased credits can therefore extend eligible paid activity without changing the usage bundled into a plan.

Sol’s input rate fell from $5 to $4 per million tokens, while cached input fell from $0.50 to $0.40. Output fell from $30 to $20 per million tokens, making generated tokens the largest component of the promotion.

GPT-5.6 Sol promotional API rates

Input
$5→$4
per million tokens

Input pricing moved from $5 to $4 per million tokens.

Cached input
$0.50→$0.40
per million tokens

Cached-input pricing moved from $0.50 to $0.40 per million tokens.

Output
$30→$20
per million tokens

Output pricing moved from $30 to $20 per million tokens.

A task using one million input tokens and producing one million output tokens would cost $24 under the temporary pricing, versus $35 previously. That example excludes tool-call fees, priority processing, cache writes and the higher rates for prompts exceeding 272,000 input tokens.

How the discount changes with the workload

  • Tasks that generate substantial output receive the largest relative benefit because output has the steepest rate reduction.
  • Workloads weighted toward new input or cached context save less than output-heavy jobs. The actual bill depends on the mix of all three token types.
  • That mix can matter in long tasks: Sol supports a 1.05 million-token context window and outputs of up to 128,000 tokens.
  • For eligible purchased-credit activity, the updated credit card lists 100 credits per million input tokens, 10 for cached input and 500 for output, compared with previous rates of 125, 12.5 and 750 credits.

OpenAI announced the reduction on August 21, and the promotion runs through at least November 21. API access is already available at the new rates, while eligible purchased-credit use in ChatGPT Work and Codex is rolling out. The company can restore the earlier rates after the promotion, so a Sol evaluation made during this period is a test of a promotional bill rather than a permanent price commitment.

Loading discussion...

Related coverage

Continue reading