We have just corrected the per-token prices listed for our text models. The prices on model pages, the docs and llms.txt were computed without a small upstream fee that billing has always included, so they read about 5.5% lower than the rate you were actually charged. The API's own /v1/ai/models listing already showed the billed rate; the pages now match it to the cent. Nothing about billing changed today and no past usage is re-billed.
What's new
- Listed per-token prices for all text models now equal the billed rate, rounded to the cent
- No charge changed: this corrects what the pages said, not what the API deducts
- Model pages, docs pages, llms.txt and /v1/ai/models now agree exactly
- Media model prices (image, video, speech, analysis) are unaffected by this correction
What changed
Example: GPT-4o Mini was listed at $0.21 / $0.84 per 1M tokens and is now listed at $0.22 / $0.89, which is what it has always cost to call. Every text model moved by the same proportion. The dashboard API Logs have always shown the real deduction per call, so your spend history is unchanged.
Why it happened
Our published price is built from the model's token rate plus our margin. The upstream credit fee that billing applies was missing from the page-side calculation, so two surfaces disagreed by 5.5%. Prices are now generated from one formula everywhere.