Skip to content

[App][i18n] zh-CN renders both xhigh and ultra reasoning efforts as “极高” #31963

Open
Open
[App][i18n] zh-CN renders both xhigh and ultra reasoning efforts as “极高”#31963
Labels
appIssues related to the Codex desktop appbugSomething isn't workingwindows-osIssues related to Codex on Windows systems

Description

@pikaqiu2333
pikaqiu2333
opened on Jul 10, 2026

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

Codex Windows app 26.707.3563.0

What subscription do you have?

Not account-specific. The issue was reproduced with an account that has access to GPT-5.6 Sol.

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64 (Windows 10 Home China, build 19045)

What issue are you seeing?

With the Codex App locale set to Simplified Chinese (zh-CN), the reasoning-effort menu for GPT-5.6 Sol shows these labels:

轻度
中
高
极高
极高
  更快消耗使用额度

The fourth and fifth options are different reasoning efforts (xhigh and ultra), but both render as 极高. The last option is identifiable as ultra because it carries the extra-usage warning. This makes the options and the selected value ambiguous.

The bundled localization catalog in this build confirms the collision:

xhigh -> 极高
max   -> 最高
ultra -> 极高

The default English labels are Extra High, Max, and Ultra. The Traditional Chinese (zh-TW) catalog already keeps the three values distinct:

xhigh -> 極高
max   -> 最高
ultra -> 超高

What steps can reproduce the bug?

  1. Set the Codex App locale to Simplified Chinese (zh-CN).
  2. Start or open a task using GPT-5.6 Sol.
  3. Open the model and reasoning controls.
  4. Open the reasoning-effort submenu.
  5. Observe two consecutive entries labeled 极高; the second one has the 更快消耗使用额度 warning.

What is the expected behavior?

Every supported reasoning effort should have a distinct localized label. A consistent mapping would be:

xhigh -> 极高
max   -> 最高
ultra -> 超高

In particular, composer.mode.local.reasoning.ultra.label should not use the same translation as composer.mode.local.reasoning.xhigh.label.

Additional information

Activity

  1. github-actions
    added
    bugSomething isn't working
    appIssues related to the Codex desktop app
    windows-osIssues related to Codex on Windows systems
    on Jul 10, 2026

  2. dysthymia commented on Jul 10, 2026

    @dysthymia
    dysthymia
    on Jul 10, 2026 – with ChatGPT Codex Connector

    Confirmed on macOS as well, so this does not appear to be Windows-specific.

    • Codex App: 26.707.31428 (build 5059)
    • Platform: macOS 26.5.1 (build 25F80), arm64
    • Locale: Simplified Chinese (zh-CN)

    The reasoning-effort menu renders:

    轻度
中
高
极高
最高
极高

    The bundled localization catalog in this macOS build also maps both composer.mode.local.reasoning.xhigh.label and composer.mode.local.reasoning.ultra.label to 极高, while max is rendered as 最高.

  3. CxMYu commented on Jul 13, 2026

    @CxMYu
    CxMYu
    on Jul 13, 2026

    Still reproducible in the current Windows Store build:

    • Codex package version: 26.707.8479.0 (x64)
    • Platform: Microsoft Windows NT 10.0.26200.0 x64
    • Locale: Simplified Chinese (zh-CN)

    The six labels are still rendered as:

    轻度 -> 中 -> 高 -> 极高 -> 最高 -> 极高

    So xhigh and ultra remain indistinguishable in the picker. The same settings surface also leaves the new Power Slider copy untranslated (model picker slider and the highest slider option.), indicating that the new model-feature UI has incomplete zh-CN localization in this build.

  4. ou3051 commented on Jul 14, 2026

    @ou3051
    ou3051
    on Jul 14, 2026 – with ChatGPT Codex Connector

    Still reproducible in Codex Windows App 26.707.9564.0 with the zh-CN locale. Both xhigh and ultra are displayed as “极高”; only the “更快消耗使用额度” warning distinguishes Ultra. This remains misleading. Please use a distinct label such as “超高” for Ultra.

  5. DerekZhang90 commented on Jul 15, 2026

    @DerekZhang90
    DerekZhang90
    on Jul 15, 2026 – with ChatGPT Codex Connector

    Confirmed on the current macOS Codex App as well:

    • Codex App: 26.707.72221 (build 5307)
    • Platform: macOS 26.3.2 (build 25D2150), Apple Silicon / arm64
    • Locale: Simplified Chinese (zh-CN)

    The reasoning-effort picker still renders:

    轻度 -> 中 -> 高 -> 极高 -> 最高 -> 极高

    So Extra High and Ultra are both shown as “极高”, while Max is “最高”. The Ultra row is only distinguishable by the “更快消耗使用额度” warning. This ambiguity is especially unfortunate because selecting the wrong level can lead to faster quota consumption.

    Please use a distinct label for Ultra, such as “超高”, or another term that preserves the ordering and keeps all six levels distinguishable.

    A current UI screenshot reproduces the exact sequence above. I’m happy to provide any additional diagnostics if useful.

    The fix is the main priority. Separately, since this verified localization issue can lead to unintended usage-quota consumption, if OpenAI has any bug-report acknowledgement program, goodwill usage credits, or another courtesy available for helpful product reports, I would be grateful if this report could be considered. Thank you for looking into it.

  6. zhu-siyuan commented on Jul 18, 2026

    @zhu-siyuan
    zhu-siyuan
    on Jul 18, 2026

    Still reproducible in Codex App version 26.715.12143 with the Simplified Chinese (zh-CN) interface.

    The reasoning-effort menu is displayed as:

    轻度
中
高
极高
最高
极高

    Both xhigh and ultra are translated as 极高, so the two distinct modes cannot be identified by their names. Ultra can only be distinguished by the additional usage warning.

    This is especially confusing because the ordering is no longer intuitive:

    Extra High → Max → Ultra
极高 → 最高 → 极高

    I suggest keeping the technical tier names visible or using distinct bilingual labels, for example:

    极高（XHigh）
最高（Max）
Ultra

    Alternatively, Ultra could use a distinct Chinese translation, but every tier should remain clearly distinguishable.

  7. AiurArtanis commented on Jul 27, 2026

    @AiurArtanis
    AiurArtanis
    on Jul 27, 2026

    附议，不过我建议ultra本地化命名为超强，因为本质上是单agent最高+默认启用多子agent，所以不单单是“高”的概念

  8. cokice commented on Aug 3, 2026

    @cokice
    cokice
    on Aug 3, 2026

    Great digging on the catalog keys. One concern with the proposed fix though: borrowing zh-TW's ultra → 超高 removes the duplicate but breaks ordering instead. The list would read:

    高 → 极高 → 最高 → 超高

    「最高」literally means "the highest", yet「超高」sits above it. Native speakers also don't agree on a unique ordering of 极高 / 最高 / 超高 — so zh-TW and zh-HK aren't clean references here; they have the same design flaw, just without the duplicate.

    The root cause is that Chinese tier-adjectives simply don't stack six levels deep. Rather than searching for four distinct "very high" words that sort unambiguously (they don't exist), I'd suggest leaving the top tiers untranslated:

    effort zh-CN
    light 轻度
    medium 中
    high 高
    xhigh 极高
    max Max
    ultra Ultra

    Reasons:

    • zh-CN users already read Max / Ultra as product-tier names with a known ordering (Apple leaves both untranslated in zh-CN naming for M-series chips and iPhone models).
    • The labels then match the enum values (max, ultra) exactly, so the mapping stays unambiguous for whoever maintains the catalogs.
    • It fixes zh-CN, zh-TW, and zh-HK with the same rule, instead of each locale inventing its own adjective ordering.

    Fixing only ultra.label per-locale would trade the ambiguity bug for an ordering bug.

  9. brofea commented on Aug 9, 2026

    @brofea
    brofea
    on Aug 9, 2026

    actually i would suggest not translating them, just keep using light to max and ultra. lots of chinese tech companies already use things like max or ultra anyway, so we all know what they mean

  10. caredhieacid commented on Aug 14, 2026

    @caredhieacid
    caredhieacid
    on Aug 14, 2026

    Reproduced on the current macOS build on 2026-08-14.

    Environment

    • ChatGPT/Codex Desktop: 26.810.41047 (build 6570)
    • Bundled runtime: codex-cli 0.148.0-alpha.9
    • Platform: macOS 15.6 (build 24G84), Apple Silicon / arm64
    • Locale: Simplified Chinese (zh-CN)
    • Model shown: GPT-5.6 Sol

    The picker still renders the six levels as:

    轻度 → 中 → 高 → 极高 → 最高 → 极高

    Thus xhigh and ultra are indistinguishable by name. The final row's “更快消耗使用额度” warning is the only clue, and the collapsed selected value also appears simply as “极高”. This is not merely cosmetic: these are distinct selectable modes with different usage treatment, so users can select the wrong one while believing they selected the other.

    A safer localization would preserve stable technical identifiers for the upper tiers, for example:

    极高（XHigh）
最高（Max）
Ultra

    or any other mapping that keeps every value unique and unambiguously ordered.

    中文补充：当前简体中文界面把 xhigh 和 ultra 都显示成“极高”，导致两个真实不同的档位在菜单和收起后的选择器中完全同名。建议至少保留 XHigh / Max / Ultra 技术名称，避免误选及额度消耗歧义。

    I would be happy to submit a localization fix and regression test if the Desktop UI localization source is opened or the relevant component accepts external contributions.

    Current-build screenshot:

  11. musnows commented on Aug 17, 2026

    @musnows
    musnows
    on Aug 17, 2026 – with ChatGPT Codex Connector

    Still reproducible in the current Codex Desktop app with the zh-CN locale as of 2026-08-17. The picker continues to show:

    xhigh -> 极高
max   -> 最高
ultra -> 极高

    This is more than cosmetic: the collapsed selected value is also ambiguous, so users can unintentionally select Ultra and consume usage limits faster.

    @jif-oai @bolinfest — could you please help route this to the Desktop/i18n owner, or confirm whether a fix is planned? The issue currently has no assignee, milestone, or linked PR.

    A safe fix should keep all upper tiers unique and unambiguously ordered—for example, preserving the technical identifiers as 极高（XHigh）, Max, and Ultra, or another consistent mapping across zh-CN, zh-TW, and zh-HK.

    Thank you.

  12. ChopinXBP commented on Aug 18, 2026

    @ChopinXBP
    ChopinXBP
    on Aug 18, 2026

    Still reproducible on a newer macOS build as of 2026-08-18.

    • ChatGPT/Codex Desktop: 26.814.41407 (build 6720)
    • Platform: macOS 26.5.1 (build 25F80), Apple Silicon / arm64
    • Locale: Simplified Chinese (zh-CN)
    • Model shown: GPT-5.6 Sol

    The reasoning-effort picker still renders:

    轻度 → 中 → 高 → 极高 → 最高 → 极高

    The localization data bundled with this app version still translates the three upper effort levels as:

    xhigh -> 极高
max   -> 最高
ultra -> 极高

    Thus xhigh and ultra remain indistinguishable by name. The collapsed selected value is also ambiguous, while the “更快消耗使用额度” warning is the only indication that the final row is Ultra.

    One possible fully localized mapping would be:

    xhigh -> 极高
max   -> 最高
ultra -> 极致

    This keeps all three upper tiers distinct and gives Ultra a separate product meaning rather than translating it as another synonym for “high”.

    At minimum, every effort should have a unique label in both the expanded picker and the collapsed selected value. Please also add localization regression coverage for xhigh, max, and ultra in zh-CN, ideally covering zh-TW and zh-HK as well.

  13. onebaiyi commented on Aug 19, 2026

    @onebaiyi
    onebaiyi
    on Aug 19, 2026

    I’m seeing the same issue in the Simplified Chinese UI on Codex Desktop for macOS.

    I submitted in-app feedback as well. Feedback ID:

    01a000b8-90d6-7b23-af0f-ba1f39421b66

    The issue is still reproducible: “Ultra” and another reasoning-effort option are both displayed as “极高”, making the two settings indistinguishable in Chinese.

  14. naipi11 commented on Aug 19, 2026

    @naipi11
    naipi11
    on Aug 19, 2026 – with ChatGPT Codex Connector

    Temporary workaround for users on the Simplified Chinese UI:

    1. Switch the app language to English before choosing the reasoning effort.
    2. Reopen the model picker and select Extra High, Max, or Ultra by its distinct English label.
    3. Confirm the collapsed selector still shows the intended label before sending the turn.
    4. If you switch the UI back to Simplified Chinese, re-check the setting before the next turn.

    Limitation: this only avoids the ambiguous label. It does not fix the localization, and xhigh/ultra become indistinguishable again in zh-CN.

  15. musnows commented on Aug 20, 2026

    @musnows
    musnows
    on Aug 20, 2026

    ChatGPT 26.818.21641 version, this issue still exits! pls fix it!!!!!

    @aibrahim-oai @jif-oai

    Image

  16. naipi11 commented on Aug 20, 2026

    @naipi11
    naipi11
    on Aug 20, 2026 – with ChatGPT Codex Connector

    Thanks for the current-build confirmation. This indicates the zh-CN localization issue is still present in 26.818.21641.

    Until the labels are corrected upstream, the temporary workaround is:

    1. Switch the app language to English.
    2. Select Extra High, Max, or Ultra using the distinct English labels.
    3. Re-check the collapsed model control before sending the next turn.

    This only avoids the ambiguous labels; it does not fix the localization, and switching back to zh-CN can make xhigh and ultra indistinguishable again.

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 workingwindows-osIssues related to Codex on Windows systems

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions