OpenCode Go subscription pricing does not match the actual charge. #39822

Activity

1. github-actions

   assigned

   StarpTech

   on Jul 31, 2026
2. John-Ferrel commented on Jul 31, 2026

   

   John-Ferrel

   on Jul 31, 2026

   Last edited by John-Ferrel

   More actions

   Have a look at
   https://opencode.ai/docs/zh-cn/go/#:~:text=%2460-,DeepSeek%20V4%20Pro,-%240.435

   Note that there's only $15 per month, so $3 per 5 hours.

   Maybe it's good idea to switch V4 Flash

   

   "For some models, we haven’t had the opportunity to negotiate a discount or host them at a lower cost, either because the model is new or because their public pricing is already discounted.

   For these models, you still get a little more than if you paid the model providers directly; this is why their usage mulitplier is lower in the table above."
3. ddd000987 commented on Jul 31, 2026

   

   ddd000987

   on Jul 31, 2026

   More actions

   I also used to think that 'only being able to use $15' meant that after I used up the $15 allowance for DeepSeek V4 Pro, I could still use other models—just not DeepSeek V4 Pro anymore.
   But in reality, it seems that usage for models capped at $15 is billed at 4x the listed price. I only realized this after seeing this discussion thread.
4. nickchomey commented on Jul 31, 2026

   

   nickchomey

   on Jul 31, 2026

   More actions

   Yes its all quite confusing. There's a good thread in reddit about it. https://www.reddit.com/r/opencode/comments/1v9cclu/opencode_gos_15_models_are_not_separate_limits/

   You have to multiply all the $15 usage model prices by 4x to be comparable to the rest. I'm sure there's no ill intent, but it's awful communication
5. BrandonKNguyen192 commented on Sep 4, 2026

   

   BrandonKNguyen192

   on Sep 4, 2026

   More actions

   Data point from a Go subscriber with three accounts, measured from the local opencode session databases (all assistant messages, Aug 28 – Sep 4, 2026):

   • DeepSeek V4 Pro on Go: 196M tokens processed (190M of them cache reads), 2.0M generated. At DeepSeek's current published list prices ($0.435 / $0.003625 / $0.87 per 1M miss / hit / output) that is about $4.26 for the week.
   • The Go dashboard shows $7.50 charged against the Pro monthly quota for Sep 1–4 alone, which puts Go's Pro accounting at roughly 3× today's list. The same window on V4 Flash (120M tokens, $1.26 at list) shows $0.01 on the dashboard.

   Practical effect: one day of Pro consumed the account's pooled weekly window while the Flash monthly quota sat at 0%. If Go still prices Pro at the pre-cut rate, adjusting the multiplier (or documenting the current one on the pricing page) would let subscribers plan usage. Happy to share the aggregation query.
6. bigprof commented on Sep 18, 2026

   

   bigprof

   on Sep 18, 2026

   More actions

   I hit the same issue and dug into the numbers — I believe I found the actual root cause, and it's not a timing/cycle mismatch (as speculated in #39430), but how the "Total" % is calculated.

   The dashboard appears to compute Total as a simple sum of each model's individual usage percentage (each model's own $ used ÷ its own $ quota), rather than a dollar-weighted aggregate (total $ used across all models ÷ total $ quota across all models).

   Example from my own weekly usage table:

   Model	Usage	Quota	%
   DeepSeek V4 Pro	$3.37	$7.50	44.9%
   GLM 5.3	$2.65	$7.50	35.3%
   GLM 5.2	$4.72	$30.00	15.7%
   Displayed Total			95.9% (= 44.9 + 35.3 + 15.7)

   But my actual dollar usage was $10.74 out of a $45.00 combined quota — about 23.9%, not 95.9%.

   The same pattern holds for my monthly and 5-hour tables — in every case, the displayed Total exactly equals the sum of the per-model percentages, not total-$-used ÷ total-$-quota.

   Because quota sizes vary a lot between models (e.g. $7.50 vs $30 vs $60), this method heavily skews the Total toward whichever models have the smallest quotas. A user could be nowhere near their real combined budget and still see a Total near 100%, which is presumably what's happening in the original report above (11% consumed off $0.35 of $12 usage).

   Could this be changed to a proper dollar-weighted aggregate (sum of $ used across all models ÷ sum of $ quota across all models)?
7. bigprof commented on Sep 20, 2026

   

   bigprof

   on Sep 20, 2026

   More actions

   Update: a day after posting the breakdown above, the Go usage page has changed — the per-model tables (with individual $ usage, $ quota, and %) are gone entirely. The dashboard now only shows a single rolled-up percentage per window (Rolling/Weekly/Monthly), with no way to see the underlying $ figures per model anymore.

   This is a step backward, not a fix. The root cause I described above (Total = sum of per-model percentages instead of a dollar-weighted aggregate) hasn't been addressed — my Weekly usage is now showing 100% with no way to verify against actual dollars spent, since that breakdown is no longer exposed. Removing the data needed to catch and verify this kind of bug, rather than fixing the calculation itself, reduces transparency for users trying to understand their own billing.

   That's a strange direction for a product built on an open-source, transparency-first project. Users should be able to see exactly how their usage is being measured, especially when there's an open, unresolved question about whether that measurement is even correct. Could the detailed per-model breakdown be restored (ideally alongside a corrected, dollar-weighted Total)?
8. thdxr commented on Oct 11, 2026

   

   thdxr

   on Oct 11, 2026

   Member

   More actions

   Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
9. thdxr

   closed this as not plannedon Oct 11, 2026