OpenAI reaches one billion active users as it cuts GPT-5.6 prices by up to 80%

midian182

Posts: 11,789   +179
Staff member
Connecting the dots: There are usually few better ways of attracting new users than dropping prices. OpenAI has just slashed what it charges for two GPT-5.6 models, a move announced one day before the company revealed that its models now reach more than one billion active users.

On July 30, OpenAI cut the price of GPT-5.6 Luna by 80%. It now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1 and $6, respectively. GPT-5.6 Terra received a smaller 20% reduction, taking its prices from $2.50 and $15 to $2 and $12. The price of Sol, OpenAI's flagship and highest-capability model, remains unchanged.

OpenAI also introduced Fast mode for Sol, which delivers up to 2.5 times the speed of standard processing at twice the price. The company says there is no reduction in intelligence when using the faster option.

The lower Luna and Terra prices are also reflected in how usage is counted against paid subscriptions in Codex and ChatGPT Work, meaning the models now consume fewer credits. Subscription prices and quota budgets remain unchanged.

OpenAI says the reductions were enabled by efficiency improvements across its models and serving systems. Within a human-led process, GPT-5.6 Sol autonomously rewrote and optimized production GPU kernels, work that helped reduce its end-to-end serving costs by 20%. Experiments designed to improve speculative decoding also increased token-generation efficiency by more than 15%.

"Better intelligence drives broader adoption. Broader adoption supports more investment. More investment improves intelligence and efficiency," wrote OpenAI CFO Sarah Friar.

Friar disclosed the next day that OpenAI's models now reach more than one billion active users and more than two million businesses. The company didn't specify whether the latest figure represents weekly or monthly users, but given its previous announcement that ChatGPT had more than 900 million weekly users, it's likely the former.

OpenAI says engagement also increases over time. Six months after signing up, people send around 50% more messages each day and use ChatGPT for about twice as many kinds of work.

All those users have yet to translate into profitability. OpenAI generated $13.07 billion in revenue during 2025 but lost $21 billion, while only around 50 million of its then-900 million weekly users paid for a subscription. It was also reported how flat-rate AI subscriptions can become unprofitable under heavy use. Lower prices could encourage even wider adoption, though analysts warn they could place further pressure on margins as the company continues spending heavily on computing infrastructure.

Permalink to story:

 
Not surprising seeing how the open source models actually providing much cheaper options with similar level of quality and accuracy. I moved from Claude to Kimi k3 and it is simply better, without stupid limitations Claude is enforcing.
 
I moved from Claude to Kimi k3 and it is simply better, without stupid limitations Claude is enforcing.
Yet here you are on another thread, with the juvenile claim that AI and datacenters mean the death of us all. Would you prefer to buy your tokens from CCP-operated datacenters, with a genocidal dictatorial regime monitoring your every request?
 
Last edited:
If stealing data to train the models is fine, stealing the outputs from the models is fine.

Unlike the data that was originally stollen, the companies that "stole" the data through prompting, at least paid the asking price. That just sounds like being a smart capitalist. Let someone else pay for the research.

I hate AI, but if you're butt hurt about China "stealing" from American companies, you also should be butt hurt about American companies stealing from the world's companies to train their models.
 
You can thank China for this.. stealing models indirectly using prompts to build thier own for a fraction of the cost.
While I don't think anyone is able to prove that China "stole" the AI models, objectively, the business of AI is pretty much a stealing business. The data that is used to train LLM models is basically obtained illegally in most cases. If they had to pay to use these content, the likes of OpenAI and Anthropic (just naming a few) would have gone bust a long time back.
 
"Anthropic told the US Senate in June 2026 that Alibaba's Qwen lab ran the largest documented model-distillation campaign on record against Claude: 28.8 million unauthorized exchanges through approximately 25,000 fake accounts. Moonshot, as an Alibaba-backed company training substantially on Alibaba's infrastructure, now faces the same accusation."
https://www.techtimes.com/articles/...dia-chips-now-qwen-losing-its-own-compute.htm

"The first time a company of Alibaba's global commercial scale has been named in a distillation complaint, and the first time Anthropic has formally submitted such evidence to the U.S. Senate"
https://www.techtimes.com/articles/...ign-against-claude-anthropic-tells-senate.htm

@J95qRP7 Both companies paid *something* (the US ones for the training cost, the Chinese for the distillation cost). And whoever were stolen from, didn't find that payment honest or sufficient. Certainly, one theft doesn't justify another.
 
Back