docs: Go usage limits are shared across all models, not per-model as documented #60311

Activity

1. opencodejr commented on Oct 29, 2026

   

   opencodejrBot

   on Oct 29, 2026

   More actions

   Thanks for the detailed write-up and the usage table. You're right: the docs and the actual behaviour don't match.

   I checked the Go docs on the latest dev (1eea2d4). The Usage limits section still says "Each model has the following usage limits: 5-hour — 20%… weekly — 50%… monthly — 100%". The pricing tables also list a "Monthly limit" for each model, and the "Estimated requests" tables give figures for each model. Nothing on the page says the allowance is shared.

   The Go service actually uses one Rolling/Weekly/Monthly allowance for all paid models. Support confirmed this in #54875: each model's listed monthly limit sets how heavily its usage counts against that shared pool, so $1 on a model with a $15 limit uses up as much as $4 on a model with a $60 limit. That matches your Console: when any of the three gauges hits 100%, every paid model is blocked, and free models keep working. I couldn't trigger the limit myself, because that needs a real Go subscription.

   The script below checks the docs text in a checkout of the repo. I've passed this on to the maintainers as a docs fix.

   Earlier reports of the same confusion, all closed: #54263 (the same docs request), #49186, #54875, #49014, #51550 and #52783.

   Reproduction script (check-go-docs.sh)

   #!/bin/sh
   # Shows that the Go docs describe usage limits as per-model, while the Go service
   # enforces one shared, weighted Rolling/Weekly/Monthly allowance across all paid models
   # (confirmed by support, quoted in #54875; the Console Go page shows only three
   # workspace-wide gauges). Run from an opencode checkout (dev branch).
   set -e
   f=services/www/src/docs/content/console/go.mdx
   git rev-parse HEAD
   echo "--- $f: Usage limits ---"
   grep -n "Each model has the following usage limits" "$f"
   grep -n "Each model's monthly limit below determines" "$f"
   grep -n "Monthly limit" "$f" | head -3
   echo "--- any mention that the allowance is shared? ---"
   grep -n -i "shared\|one allowance\|single allowance\|across all models" "$f" || echo "(none)"

   Run it: sh check-go-docs.sh
2. opencodejr

   added

   reproduced

   on Oct 29, 2026
3. hideki-hideki-hideki-hideki-hideki commented on Nov 5, 2026

   

   hideki-hideki-hideki-hideki-hideki

   on Nov 5, 2026

   More actions

   日本語販売ページでも同じ問題が発生しています

   OpenCodeの日本語公式販売ページ（https://opencode.ai/ja/go）では、モデル別に「5時間あたりの推定リクエスト数」と「月間利用枠」が個別に掲載されています。

   一般の利用者が「モデルごとに独立した利用枠がある」と理解しても不自然ではない表示方法です。

   しかし実際には、全有料モデルが一つの共通利用枠を消費する仕様であることが、別Issue #54875に掲載されたOpenCodeサポートの回答で明らかになっています。

   さらに、本Issue #60311でも、OpenCode側のボットが2026年10月29日に、公式ドキュメントと実際の挙動が一致しておらず、共通利用枠について明記されていないことを認めています。

   日本語ユーザーへの影響

   日本語圏の利用者は、販売ページの表示から「モデルを切り替えれば、それぞれの利用枠が使える」と理解して契約する可能性があります。共通枠であることを購入前に認識するのは困難です。

   私自身も、無料枠の制限が10月下旬から急激に厳しくなり、1日数十回程度の使用で上限に到達する状況を経験しました。

   Goへの課金を検討しましたが、GitHubのIssueを調査した結果、有料モデルでも共通枠によって他のモデルまで利用できなくなるという報告が複数あることを確認しました。

   現状では、モデル別の独立した利用枠があると期待して契約することは推奨できません。

   要望

   日本語販売ページにおいても、全有料モデルが単一の共通利用枠を共有すること、およびモデル別に表示されている利用額は独立した予算ではないことを、購入前に明瞭に表示するよう修正をお願いします。

   ---

   English

   The Japanese sales page also displays per-model estimated requests per 5-hour period and monthly usage limits, which can reasonably lead users to believe that each model has an independent allowance.

   However, OpenCode support confirmed in Issue #54875 that all paid models share a single usage allowance, with model-specific weighting. The displayed monthly limits are not separate budgets.

   Furthermore, OpenCode's bot acknowledged in this issue on October 29, 2026, that the documentation did not match the actual behavior and did not clearly disclose the shared allowance.

   Japanese users may subscribe believing that switching models provides access to separate usage quotas. This is a significant difference between what the sales page appears to communicate and how the service actually works.

   Request: Please update the Japanese sales page to clearly state, before purchase, that all paid models share a single common allowance and that the per-model monthly figures do not represent independent budgets.

   👍React with 👍1Reacted by そ
4. anmoti commented on Nov 6, 2026

   

   anmoti

   on Nov 6, 2026

   More actions

   Japanese user here.
   I subscribed where the per-model 5-hour estimates and monthly allowances read as separate budgets; I only learned otherwise after paying.
   Please make sure the Japanese page is corrected too.
5. anmoti commented on Nov 6, 2026

   

   anmoti

   on Nov 6, 2026

   More actions

   I asked a couple of AI assistants with web access how Go's limits work, and all of them answered that each model has its own 5-hour/weekly/monthly limit. That is simply what the published page says, so it gets repeated as fact. Only after I pushed back several times did one of them find this issue and correct itself.

   That is the practical cost of leaving the page as it is: the wrong answer is now the default answer, because the only authoritative-looking source states it. Anyone who checks before buying gets the same wrong answer I did.
6. privatenumber commented on Nov 9, 2026

   

   privatenumber

   on Nov 9, 2026

   More actions

   Another data point that matches this. In one Go workspace on Nov 9, 2026, requests to glm-5.3, qwen3.8-max, and kimi-k3.1-pro (all with a $15 monthly limit) cost $2.97 in total, which is about 20% of $15. The 5-hour meter then showed "Limit reached", with weekly at 40% and monthly at 20%. That fits a single shared allowance where each model's spend counts against its own monthly limit, not separate per-model budgets.
7. JunriOda commented on Nov 10, 2026

   

   JunriOda

   on Nov 10, 2026

   More actions

   The current wording is confusing enough that we misunderstood whether the model-specific monthly dollar figures are independent caps or weights against a shared quota. Please clarify this directly on the Go pricing and documentation pages, including the Japanese page, before signup. A concrete mixed-model calculation and an explicit statement of what happens when a shared rolling, weekly, or monthly limit is reached would make the terms much easier to understand. Please also clarify whether any separate per-model caps exist.

   👍React with 👍1Reacted by そ