Just saw the announcement. The new "tiered" pricing looks like classic vendor lock-in dressed up as a favor.
Ran the numbers for our monthly usage (~12M tokens across GPT-4o and o1). The so-called discount is a rounding error unless you're a mega-corp.
Key issues:
* The commitment trap. Lock in a tier, your usage creeps up, you're suddenly paying for the next tier anyway.
* Ignores the real cost: reliability. Their tiered support is the actual price tag. Good luck getting help on the "Standard" tier when an endpoint goes sideways during a campaign launch.
* Doesn't touch the cost of fine-tuned models or the upcoming "reasoning" credits. That's where the real bill lives.
Anyone else seeing a path where this actually moves the needle on your unit economics? Or is this just a psychological play to make us feel like we're "negotiating"?
I'm a principal engineer at a mid-sized e-commerce platform, leading our AI product team. We run a mix of GPT-4o for live chat and o1-preview for search query refinement in production, processing around 15M tokens per month.
* **Target Audience - Enterprise Procurement Only:** The tiered pricing is structured for finance teams, not engineers. The discount curve is negligible below ~$5-10k/month in committed spend. For a company spending $2k/month, a 2-3% discount is just noise; the real value is for organizations already spending six figures, where they might see 10-15% off list.
* **Hidden Cost is Support Tiering:** Your effective price is list minus discount plus the cost of adequate support. We're on the "Scale" support tier ($4,500/month) because we need <4-hour response for P1 issues. On "Standard" support (included), our tickets during an o1 outage last month took over 18 hours for a first response. The real cost of these tiers is the support add-on you'll need to buy.
* **Deployment Lock-In is Minimal:** The commitment is purely financial, not technical. You're committing to a spend level, not re-architecting. The trap is behavioral: your team will naturally increase usage to meet the committed amount, which is a well-documented cloud cost phenomenon. We saw this with AWS Reserved Instances; our usage grew 30% the quarter after we bought them.
* **It Ignores High-Cost Services:** The tier discount only applies to base model usage (Chat Completions, Completions API). It does not apply to fine-tuning jobs, Batch API costs, or the new reasoning credits. Our fine-tuning jobs for a specialized classifier are 40% of our OpenAI bill and are completely excluded from the tier calculation, which dramatically reduces the overall impact.
Given your stated ~12M token usage, the discount is likely under 2%. I wouldn't recommend the commitment unless your finance team requires predictable billing. The only scenario where this makes sense is if you're already spending over $8k/month on base models *and* your usage growth is predictable and flat. To make a clean call, tell us your current monthly spend on base models alone and whether you use fine-tuning.
Less spend, more headroom.