OpenCode Go weekly/monthly quota does not match usage history #41206

Activity

1. github-actions commented on Aug 8, 2026

   

   github-actionsBot

   on Aug 8, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Go quota shows 100% exhausted but visible usage is only ~$20 #38936: Go quota shows 100% exhausted but visible usage is only ~$20 — same symptom of visible usage not matching quota percentage
   • Incorrect Quota in Opencode GO consumption #40031: Incorrect Quota in Opencode GO consumption — same discrepancy between actual spend and reported quota consumption

   Please review those issues and add any additional information there if they cover your situation.
2. aliaksandrsen commented on Aug 9, 2026

   

   aliaksandrsen

   on Aug 9, 2026

   Last edited by aliaksandrsen

   More actions

   Request for clarification: how model usage maps to Go quota limits

   Thank you for investigating this. I would like to clarify the accounting model behind the Go limits, because the numbers in this issue appear to show that the dollar amounts in Usage History are not the same dollar amounts used by the rolling, weekly, and monthly quota counters.

   According to the OpenCode Go documentation, the plan limits are:

   • 5-hour limit: $12 of usage
   • Weekly limit: $30 of usage
   • Monthly limit: $60 of usage

   The documentation says that these limits are defined in dollar value, and that the number of requests depends on the model. It also provides an estimated request-count table for each model. However, it is not clear whether the dollar amounts shown in the Usage History are the exact dollar values deducted from these quota windows, or whether Go applies a separate internal normalization, multiplier, model weight, or other accounting rule.

   Numbers from this issue

   The Usage History in this report shows approximately:

   • DeepSeek V4 Pro: $2.01
   • GPT-5.6 Luna: $6.46
   • DeepSeek V4 Pro: $0.04
   • Qwen3.8 Max: $2.58
   • Total visible Go usage: approximately $11.09

   The quota dashboard simultaneously shows:

   • Rolling usage: 87%
   • Weekly usage: 100%
   • Monthly usage: 52%

   If the visible Usage History amounts were used directly, $11.09 would correspond to approximately:

   • $11.09 / $30 = 37% of the weekly limit, not 100%
   • $11.09 / $60 = 18.5% of the monthly limit, not 52%

   The monthly value of 52% corresponds to approximately $31.20 of internal monthly quota usage, despite only $11.09 being visible in Usage History. The weekly value shows an even larger discrepancy, because the weekly window is already exhausted.

   Related concrete examples

   Other reports show the same pattern:

   • Issue #41146 reports approximately $7.52 of visible qwen3.8-max usage, while the $30 weekly limit is at 100% and the $60 monthly limit is at 50%.
   • Issue #40031 reports approximately $3 of visible Grok usage, while the session usage is shown as 100% of a $12 rolling limit.
   • Issue #38936 reports approximately $20 of visible usage, while the $60 monthly quota is shown as fully exhausted.

   These examples suggest that the quota system may be charging an internal usage amount that is different from the user-visible Usage History amount. They also suggest that the relationship may be model-dependent.

   Direct questions about the model table and the shared quota

   Could the OpenCode team please answer these questions explicitly?

   1. Is the dollar amount shown for a request in Usage History the same dollar amount that is added to the Go rolling, weekly, and monthly quota counters?
   2. If not, what is the exact formula used to convert a request into quota usage?
   3. Are there model-specific multipliers, weights, minimum charges, context-length adjustments, prompt-cache adjustments, retries, or provider-side charges that are included in quota accounting but omitted from Usage History?
   4. Are the 5-hour, weekly, and monthly counters based on one shared normalized usage value, or are they calculated independently?
   5. When the documentation or model table shows a model Usage value of $15 while the Go monthly limit is $60, what exactly does that mean?
   6. For example, if a user sends requests to Grok 4.5 and the visible Usage History total for Grok reaches $15, does that:
      • consume $15 of the shared $60 monthly Go quota, leaving approximately $45 for other models; or
      • consume the full $60 monthly quota because Grok has an effective 4x multiplier; or
      • follow some other model-specific calculation?
   7. Conversely, if another model also has a displayed Usage value of $15, does it consume the same amount of shared quota as Grok, or can two models with the same displayed value have different quota weights?
   8. Does the Usage column represent a per-model allowance, an estimated raw provider cost, an internal quota-equivalent amount, or merely an estimate used to derive request counts?
   9. Can you provide one worked example showing input tokens, output tokens, cache state, displayed Usage History cost, normalized Go quota cost, and the resulting percentage of each quota window?

   A clear explanation of this distinction is important because, based on the current UI and documentation, a reasonable user would interpret $15 of model usage as consuming $15 from the shared $60 monthly allowance. The reports above instead appear to indicate that the quota can be consumed several times faster than the visible dollar usage suggests.

   If the behavior in these reports is intentional, please document the formula and the meaning of the model table. If it is not intentional, please confirm whether these are quota-accounting bugs and whether affected accounts will be corrected.

   

   https://opencode.ai/docs/en/go/#usage-limits:~:text=The%20estimates%20are%20also%20based%20on%20the%20following%20prices%20per%201M%20tokens%20and%20the%20monthly%20usage%20included%20with%20each%20model%3A
3. aliaksandrsen commented on Aug 9, 2026

   

   aliaksandrsen

   on Aug 9, 2026

   More actions

   I opened a separate issue for the broader quota-accounting question, including the documentation link, calculations, related reports, and direct questions about model multipliers and the meaning of the $15 Usage values:

   #41391

   This issue can remain focused on the concrete account-level mismatch described here.
4. SubsUser commented on Aug 27, 2026

   

   SubsUser

   on Aug 27, 2026

   Last edited by SubsUser

   More actions

   i also have the same issue. here are some screenshots from my workspace wrk_01KYACBE50Q11B3AMGQ47XDP83

   

   i only spent 0.15 $ and reached 100% usage. please fix my bug and give me refund, i paid for a service that i cant use.
5. zlymeda commented on Aug 29, 2026

   

   zlymeda

   on Aug 29, 2026

   More actions

   same issue here:

   

   the percentage calculations are wrong! it is just summing the lines, which is incorrect.
   weekly usage is $9.4691 out of $30 (weekly limit) which is 31%, but it shows 100%!

   seems like if you used 1 model eg $15 out of $15 it would say 100%! but the weekly limit is $30.
6. peqwilber commented on Sep 25, 2026

   

   peqwilber

   on Sep 25, 2026

   Last edited by peqwilber

   More actions

   

   I'm having the same issue with OpenCode Go.
   My weekly usage is showing as exhausted, but my Usage History for the last 7 days shows only about $21.82 total cost.
   My usage includes:
   DeepSeek V4.1 Flash: $6.64
   MiMo V2.6 Flash: $5.88
   GLM 5.3 Flash: $3.79
   DeepSeek V4 Pro: $2.07
   Kimi K3: $1.87
   Grok 4.6: $0.88
   MiMo V2.5: $0.50
   Muse Spark 1.3: $0.14
   MiniMax M3: $0.05
   Total visible cost: approximately $21.82, while the documented weekly Go limit is $30.
   Could you please clarify how my weekly quota was calculated and whether the account quota is being applied correctly?
7. alkaid-ops commented on Sep 27, 2026

   

   alkaid-ops

   on Sep 27, 2026

   Last edited by alkaid-ops

   More actions

   Quite unauthentic

   $0.38 ÷ 3% ≈$12.7 Well below the advertised $60

   9.29 edit
   Actually, no problem. The biggest problem is the documentation.
   I finally understand the opencode go quota calculation. The expensive model is calculated based on the actual price of 10 US dollars you recharge. The cheapest model is calculated as 60 US dollars with a 1.4% discount. However, this discount is not displayed on the official website.
8. NEWNEWLILI commented on Oct 14, 2026

   

   NEWNEWLILI

   on Oct 14, 2026

   More actions

   Adding another concrete data point for an affected account (Go, $10/month, single member).

   Snapshot at 2026-10-14T03:05Z (/zen/go/v1/usage):

   • rolling (5h): 22%, resets 2026-10-14T05:58:18Z
   • weekly: 40%, resets 2026-10-19T00:00:00Z
   • monthly: 43%, resets 2026-11-11T02:52:40Z

   So the monthly window started at 2026-10-12T02:52:40Z, and the weekly window started at 2026-10-12T00:00:00Z. That means both windows cover essentially the same span (Oct 12 → now, ~2.5 days).

   Visible usage for that same span, from the console export (/api/v2/usage/export):

   UTC day	visible cost
   2026-10-12	$4.98
   2026-10-13	$5.56
   2026-10-14 (partial)	$2.38
   total	≈ $12.9

   All of it is DeepSeek V4.1 Flash, and those dollar amounts already include the peak/off-peak pricing (I verified: weekend days come out at ~1.0× my local off-peak estimate, weekdays at ~1.3–1.8×).

   Now the inconsistency:

   • Weekly = 40%. At the documented $30 weekly limit that implies ≈ $12.0 of usage — consistent with the ≈ $12.9 visible.
   • Monthly = 43%. At the documented $60 monthly limit that implies ≈ $25.8 of usage — about 2× the visible ≈ $12.9.

   Because the weekly and monthly windows span the same period, the same underlying usage should make the weekly percentage double the monthly percentage (≈ 43% weekly vs ≈ 21% monthly on $30/$60). Instead the dashboard shows 40% and 43% — nearly equal.

   That leaves two possibilities: the monthly counter is charging roughly 2× the visible dollars, or the monthly denominator is ~$30 rather than the documented $60.

   Questions:

   1. What is the exact formula that converts visible usage dollars into each quota window?
   2. Why do the weekly and monthly windows — which cover the same period — imply different total usage?
   3. Is this the broader quota-accounting issue (Go quota accounting does not match Usage History or documented limits #41391)? Is a fix or account credit planned?

   Environment: OpenCode TUI on Linux, model opencode-go/deepseek-v4.1-flash, ordinary coding-agent traffic (tool-calling sessions). I can provide the raw CSV export if useful.
9. Tarquinen commented on Oct 20, 2026

   

   Tarquinen

   on Oct 20, 2026

   Member

   More actions

   We're tracking this in #42985, so I'm closing this as a duplicate. Please follow that issue for updates.

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
10. Tarquinen

    closed this as a duplicateof OpenCode Go quota usage appears ~4x higher than displayed DeepSeek V4 Flash cost #42985 on Oct 20, 2026