Skip to content

GPT-6 Astra: Usage limits consuming rapidly / short session length #43201

Open
Open
GPT-6 Astra: Usage limits consuming rapidly / short session length#43201
Labels
appIssues related to the Codex desktop appbugSomething isn't workingrate-limitsIssues related to rate limits, quotas, and token usage reporting

Description

@imtiyazali73
imtiyazali73
opened on Sep 6, 2026

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.901.51231

What subscription do you have?

Plus

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

I am writing to share feedback regarding the usage limits for the new GPT-6 Astra model. Currently, the rate limit is consumed far too quickly, making it nearly impossible to complete even minor tasks within a single session. 
For context, I started a fresh session with 100% capacity at 2:51 PM IST, and by 3:11 PM IST, I had already hit the limit. A 20-minute window is simply too short to accomplish meaningful work. Having to constantly pause and wait for the next session creates a highly disruptive and frustrating workflow. 
I strongly recommend reviewing the current limits for GPT-6 Astra, as these brief sessions significantly diminish the model's practical utility.

What steps can reproduce the bug?

  1. Open a fresh session with the GPT-6 Astra model showing 100% availability (Started at 2:51 PM IST).
  2. Begin working on standard development/text tasks.
  3. Observe that the session limit is completely exhausted within 20 minutes (Session went out of limit at 3:11 PM IST).
  4. The rapid consumption happens even during normal, small-scale usage without massive context inputs or heavy code generation.

What is the expected behavior?

The GPT-6 Astra session limit should be structured to allow for reasonable, continuous workflow chunks. A single session should last long enough to complete at least minor-to-moderate tasks (ideally extending beyond a 20-minute window) without abruptly cutting off and forcing the user into long wait times for the next session.

Additional information

This issue heavily impacts productivity. Having to repeatedly pause work and wait for the cooldown/refresh period creates a disjointed experience and introduces unnecessary stress when trying to maintain a steady workflow.

Activity

  1. imtiyazali73
    added
    appIssues related to the Codex desktop app
    on Sep 6, 2026
  2. github-actions
    added
    bugSomething isn't working
    rate-limitsIssues related to rate limits, quotas, and token usage reporting
    on Sep 6, 2026

  3. github-actions commented on Sep 6, 2026

    @github-actions
    github-actionsbot
    on Sep 6, 2026 – with GitHub Actions
    Contributor

    Potential duplicates detected. Please review them and close your issue if it is a duplicate.

    Powered by Codex Action

  4. Suh0161 commented on Sep 6, 2026

    @Suh0161
    Suh0161
    on Sep 6, 2026

    I’m also seeing unexpectedly rapid usage consumption on Windows with Astra Light / low reasoning. I’m on the ChatGPT Business plan.

    Environment:

    • Subscription: ChatGPT Business
    • Model: gpt-6-astra
    • Reasoning effort: low (Light)
    • Service tier: default
    • Platform: Windows
    • Execution: Local
    • Codex App version: 26.901.51231
    • Approval policy: never
    • Sandbox: full/elevated access
    • Memories: disabled

    Observed behavior:

    • Usage before: ~98% remaining
    • Number of prompts/tasks: 2
    • Usage after: 0% remaining

    The tasks were ordinary local repository work: auditing one local repository and then working with another local repository. They were not unusually long-running workloads and did not involve intentionally huge context inputs.

    Relevant sanitized config:

    model = "gpt-6-astra"
model_reasoning_effort = "low"
service_tier = "default"

approval_policy = "never"
sandbox_mode = "danger-full-access"

[windows]
sandbox = "elevated"

[features]
memories = false
chronicle = false
js_repl = false

  5. alexf3791 commented on Sep 7, 2026

    @alexf3791
    alexf3791
    on Sep 7, 2026

    I can reproduce what appears to be the same issue, on the exact same Codex Desktop version.

    Environment:

    • ChatGPT Plus
    • Windows
    • Codex Desktop 26.901.51231
    • GPT-6 Astra
    • Reasoning: Very High

    A single task lasting 11m31s consumed:

    • 5-hour allowance: 100% -> 0%
    • Weekly allowance: 100% -> 83%

    The task involved normal code changes/commands, edited 7 files, and had one automatic context compaction. No other Codex or ChatGPT Work tasks were running.

    I submitted diagnostic feedback from the affected session.

    Feedback ID:
    01a0781e-c5b3-7110-bf8a-2e564861a24b

  6. mshata96 commented on Sep 8, 2026

    @mshata96
    mshata96
    on Sep 8, 2026 – with ChatGPT Codex Connector

    I’m seeing the same problem, and the consumption looks clearly disproportionate.

    During one ongoing Codex task using GPT-6 Astra, I used a usage reset in the middle of the task. The reset brought my allowance back to 100%. I then continued the exact same task, and before the task was even finished, my remaining allowance had already fallen below 70%.

    So more than 30% of a freshly reset allowance was consumed by continuing a single unfinished task. This was not a new project or a series of separate heavy jobs; it was one continuous task.

    I understand Astra may have a higher compute/quota cost than other models, but this rate of consumption does not feel reasonable or predictable for normal development work. Please investigate whether there is an issue with Astra usage accounting, context/reprocessing overhead, sub-agent/tool activity, retries, or the quota weighting applied to the model.

    It would also be very helpful to expose clearer per-task/per-turn usage information so users can see what is actually consuming the allowance.

  7. robert-lukowski commented on Sep 8, 2026

    @robert-lukowski
    robert-lukowski
    on Sep 8, 2026

    I’ve noticed similar Astra usage behavior in my own engineering tasks.
    Instead of treating this only as a quota problem, I started looking at how the workload was being assigned to the model. A significant part of the context and execution did not actually require premium reasoning.
    I’ve been experimenting with splitting the workflow across specialized agents/models:
    Router → Context Scout → Premium Reasoning → Implementer → Reviewer
    Repository discovery and context filtering are handled before the expensive reasoning step. Routine implementation, tests, documentation and Git operations can then be delegated to smaller models, while Astra is reserved for architecture, security and difficult reasoning.
    This doesn’t change Astra’s underlying quota consumption, but it can reduce how much unnecessary work is sent to it.
    I turned the experiment into a small open-source GitHub template:
    https://github.com/robert-lukowski/token-gourmand
    Still early, and I’m planning to benchmark the actual reduction rather than assume the savings.

  8. gus6443-beep commented on Sep 11, 2026

    @gus6443-beep
    gus6443-beep
    on Sep 11, 2026 – with ChatGPT Codex Connector

    Plus subscriber here. My weekly Astra allowance was exhausted after roughly five image-generation / Work tasks. The issue is no longer just that Astra consumes quota faster than Sol — the available amount is too small to complete normal Work workflows reliably.

    I would strongly prefer either a meaningful increase to the Plus Astra allowance or an intermediate plan between Plus and Pro (for example, a 2.5x–5x usage tier). The current jump from Plus to Pro is too large for users who need more Work/Astra capacity but do not need the highest tier.

    More broadly, I think the product would be easier to use if Astra's quota consumption were made more predictable and if long-running Work tasks could complete without being interrupted by a hard usage-limit screen mid-flow.

  9. bilalhussainx commented on Sep 11, 2026

    @bilalhussainx
    bilalhussainx
    on Sep 11, 2026

    I switched from Max Anthropic Plan thinking I could get more usage from a comparable frontier model as Fable 5.1 (GPT 6-Astra) and I got the expensive 5x Pro Use membership and GPT 6 Astra created goals that it continued to work towards for 3-4 hours consuming 65-70% of my weekly limit from one session. Way to lose a new user.

  10. D4ttara commented on Sep 15, 2026

    @D4ttara
    D4ttara
    on Sep 15, 2026 – with ChatGPT Codex Connector

    I have a closely matching Astra data point from ChatGPT Plus with local Codex credit-balance telemetry.

    On Sep 9:

    • ~23:19:50 GMT+2: purchased-credit balance = 250.0000
    • session model: GPT-6 Astra
    • reasoning effort: High
    • ~23:42:32: balance = 4.240498
    • shortly afterward: balance = 0

    So nearly 250 purchased credits were consumed in roughly 23 minutes.

    This is strikingly close to the ~20-minute session-length behavior reported in this issue.

    I have not found evidence so far that this specific rollout spawned a large subagent swarm, so I cannot explain the burn that way. The session does contain substantial long-context/compaction activity, but I am not assuming that explains billing without a proper breakdown.

    The practical problem is that the current Analytics UI shows aggregate credit spend but does not provide enough task/turn-level attribution to determine what actually drove the rapid Astra consumption. OpenAI Support has told me they cannot provide per-task server-side ledger attribution by session/thread.

    It would be very useful for this issue if OpenAI could clarify whether this rate is expected for Astra / High and expose the metering components responsible for it.

  11. f1ames0ff commented on Oct 10, 2026

    @f1ames0ff
    f1ames0ff
    on Oct 10, 2026

    GPT pricing is an absolutely unpredictable shit. Use DeepSeek instead.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    appIssues related to the Codex desktop appbugSomething isn't workingrate-limitsIssues related to rate limits, quotas, and token usage reporting

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions