OpenAI Just Made Its Best Model a Third Cheaper, and Claude Opus 5 Is Suddenly the Pricier Option
OpenAI cut GPT-5.6 Sol's API prices by 20 percent on input and 33 percent on output, guaranteed through November 21. It is the second cut in a month, and it now undercuts Anthropic's flagship on both ends.
OpenAI has cut the API price of GPT-5.6 Sol, its most capable model, for the second time in a month. Since August 21, developers pay $4 instead of $5 per million input tokens, and $20 instead of $30 per million output tokens, a 20 percent cut on what you feed the model and a 33 percent cut on what it writes back. OpenAI says the new rate holds through at least November 21.
Sol launched on July 9 at the old prices while its smaller siblings, Terra and Luna, got their own price cuts on July 30. Sol had stayed put until now, making it the most expensive model in OpenAI’s lineup. The company hasn’t said what happens after November, so treat this as a promotional window rather than a permanent price cut.
What actually changed, in plain terms: an API price is what a developer or a business pays per “token,” roughly a chunk of a word, to send text to the model (input) and to receive its answer (output). A workload that used to cost $45 for a mix of input and output tokens now costs about $32, a saving close to 29 percent. Cached input, meaning text the model has already seen recently and doesn’t need to reprocess from scratch, gets the same 20 percent discount. If you use ChatGPT through a paid subscription rather than the API, nothing changes for you: this cut only applies to metered API use, Codex credits, and eligible ChatGPT Work plans, not to what’s already included in Pro, Plus, or Business.
Why now: the timing lines up neatly with Anthropic’s own pricing. Sol’s new rate puts it below Claude Opus 5 on both input and output cost, after months where OpenAI’s flagship was the pricier of the two. Frontier labs rarely cut prices out of generosity. This reads as a direct response to competitive pressure, the kind of move you make when a rival’s model is winning API customers on cost rather than capability.
What this means for you: if you’re a developer or a small business paying per token through the API, this is a straightforward win, especially if your workload leans heavily on generated output (code, long answers, documents), where the savings are biggest. If you’re a regular ChatGPT subscriber, this doesn’t touch your bill at all. And if you’re choosing between OpenAI and Anthropic for a new project, it’s worth re-running your cost estimate now that the two flagships have effectively swapped places on price.
Sources
A Support Bot Went From Solving a Quarter of Tickets to Solving Half, Without a Better Model
Pinecone's Nexus knowledge layer is now generally available. In its own test, the same AI models cut their cost per task by up to 80 percent and roughly doubled a support agent's resolution rate, simply by giving them better-organized company knowledge instead of a smarter brain.
