Facing Rate Limit Issues with o4-mini-deep-research Model

API
, , ,
1

We are currently working with the Open AI o4-mini-deep-research model using the Agents SDK as outlined in this example: Introduction to Deep Research API Agents. However, we are frequently encountering the following error:

Rate limit reached for o4-mini-deep-research in organization on tokens per min (TPM): Limit 200000, Used 177684, Requested 33606. Please try again in 3.387s.

To address this, we implemented an exponential backoff retry, but it hasn’t resolved the issue. Even when we retry the API call after a few minutes, we still face the same rate limit problem.
Here are some details:

  • We are currently on Tier-1.
  • We are open to upgrading to a higher tier, but I couldn’t find any documentation confirming that moving to a higher tier would increase the TPM limit for this specific model (currently set to 200,000 by OpenAI).

Additionally, we wanted to try implementing this with the o3-deep-research model, but I don’t see it listed under my OpenAI subscription. Currently, I can only see two deep research models available i.e. o4-mini-deep-research and o4-mini-deep-research-2025-06-26

Is o3-deep-research restricted to certain subscription tiers, or is there another reason it might not be accessible?

I’ve attached relevant screenshots for additional context. Has anyone faced similar issues or have insights on how to handle this?

Would upgrading the tier resolve the token-per-minute limit for o4-mini-deep-research, and how can we access the o3-deep-research model?

Any guidance would be greatly appreciated! Thank you.

2

hey there - may I ask how you managed to implement a backoff inside the agent SDK run function? I cant seem to figure out where to put it. Thanks in advance!

3

Here you go:

Tier o3 Deep Research (TPM) o4-mini Deep Research (TPM)
Tier 1 200,000 200,000
Tier 2 450,000 2,000,000
Tier 3 800,000 4,000,000
Tier 4 2,000,000 10,000,000
Tier 5 30,000,000 150,000,000

Tier 2 is gained by:

  • initial payment over 7 days ago
  • newest payment brings total paid to OpenAI over $50

o3 requires “ID verification”: scan your personal ID and video your face and send it off to a sketchy company, in your platform site “general” section. It seems that ID verify requirement is inherited by the o3 deep research version, also (which can be costly - some new changes should make it closer to $2 per call instead of $20.)

4

This is just stupid and disgusting.

One cannot make ANY meaningful DR with tier 1 limit, and the only way to bypass that is waste $50 on garbage?

Fwiw, I use gemini and claude for other API calls.

Our only use case for openai is the DR, which is entirely useless without throwing money down the drain with nothing in return.

Why topping up credits is not sufficient??

1 Like
5

I’m having the same issue (and am also unwilling to waste money just to be able to spend more money). Did you find a solution to this?

6

Using Deep Research will cost you a lot of money. It is expensive—a single use can be several dollars.

Let’s say that your first payment is older than two weeks, and you now pay in enough that your total payments to OpenAI are over $100, so you’ve reached tier 3, with TPM=800,000 (on a model that can cost you into the millions per call at $10/M in and $40/M out, because it can “research” for dozens and dozens of tool calls).

How many uses are you actually going to get for your balance you’ve paid in? Not a whole lot. Compare that to $200/month for ChatGPT Pro, which gives you 250/month…where you find you don’t actually have that much to “research” anymore after paying $2,400 in month 13.

7

The thing is I don’t have any costs racked up on my account right now because apart from deep research, I am only using very light models. And I can’t do any deep research calls because they error out due to rate limits.

Pro is not an interesting option for me as I am specifically looking for API solutions

8

The “costs racked up”, or, how much you have “spent” or “consumed”, are not considered in tier elevation.

Only:

  • time since the first payment;
  • total cumulative amount paid at the time of the latest payment, inclusive.

So If you paid $10 two weeks ago, and add $41 (just to be safe) - you get tier 2.

Then you have $50 of credit balance to spend in the next year before expiry on API AI products.

9

Sorry for the confusion, took me a while to understand but I got it now. I am contacting support to refresh my usage tier based on the amount already in my wallet. Thanks for the help

1 Like
10

The language is extremely confusing. If what’s being described is the case the language should be as follows:

  • At least $100 spent on the API or added to the API since account creation.

Further despite that I have another account topped up, created a month ago, that account is still tier 1.

image
image506×160 6.42 KB

image
image1026×330 30.3 KB

11

Yes, the language is wrong in the platform site and must be fixed. A version without mistruth:

Your tier level is recalculated at the time of making a new credit balance purchase, based on these criteria:

  • the time between the organization’s initial credit purchase and now,
  • the cumulative amount then paid to OpenAI to purchase API credits.

You must interact with the payment system to receive any tier re-evaluation by making a new credit purchase; it is never automatic.

12

did you ever get a work around for this?? Anybody who has??