one year on
OpenAI unveils o1 series models with hidden chain-of-thought reasoning
The new models, which the company says spend more time 'thinking' before responding, are priced at $15 per million input tokens and $60 per million output tokens, with the chain-of-thought tokens billed as output even when invisible to the user.
OpenAI today released its new o1 series of models, which it describes as spending more time ‘thinking’ before responding to complex queries. The announcement, made via a blog post and API documentation, introduces two models: o1-preview and o1-mini.
The o1-preview model costs $15 per million input tokens and $60 per million output tokens, with a context window of 128,000 tokens and a maximum output of 32,768 tokens. The o1-mini, priced 80% cheaper at $3 and $12 per million tokens respectively, doubles the maximum output to 65,536 tokens. Access to the models via the API requires a tier 5 user level — meaning $1,000 total paid and 30+ days since first successful payment — while ChatGPT Plus users have access in-app.
A key point of debate is the hidden chain-of-thought reasoning. The API bills these internal reasoning tokens as output tokens even though they are not visible to the user. The blog cites ‘competitive advantage’ among reasons for keeping the reasoning hidden, a decision that drew criticism from developers concerned about opaque billing and inability to verify costs. Some commenters drew analogies to ‘the world’s most scrutinized employee’ while others worried about the practicality of cost control in deployed products.
On Hacker News, the thread attracted 1,654 points and 1,261 comments, with the community extensively dissecting pricing, access restrictions, and the implications of hidden reasoning tokens.
Flagged tier-5 API access and pricing, but said it was not clear whether hidden chain-of-thought reasoning is billed as paid output tokens. Also noted that the specialized coding version mentioned in the blog is not yet available.
Confirmed that reasoning tokens are billed as output tokens, citing the OpenAI docs.
Pointed out that the decision to hide chain-of-thought likely stems from competitive advantage concerns, as visible CoT could be copied.
Criticized the explanation for hiding CoT as 'a bad explanation' and noted the real reason is likely competitive advantage.
Expressed skepticism about the accuracy graphs lacking time labels, suggesting cherry-picking of data.
One year later — open only if you can handle spoilers
The o1 series marked a major shift in OpenAI's product strategy toward reasoning-focused models. Within a year, similar 'thinking' models became standard across the industry, and the debate over hidden chain-of-thought billing set a precedent for opaque pricing that persisted in subsequent releases.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy