Release notes

🛠️ Enhancements

🔤 Autocorrect

• Improved autocorrect handling in email addresses. Autocorrect is now disabled when a word is part of an email (e.g., typing brian@... will no longer autocorrect brian to Brian). In other contexts, capitalization suggestions remain active.

📝 TL;DR

• Major dictionary expansion for English, German, and Spanish.
• New cross-language database of 461 proper names and 212 abbreviations.
• Extended style guides with inclusive language (English) and 1,624 new Ukrainian entries.
• Cleaner autocorrect with invalid pairs removed and smarter prioritization for typos and names.
• Grammar engine updates for German and Spanish with better punctuation, case sensitivity, and compounds.
• New AI proofreading engine for English with fixes for dashes, pluralization, medical terms, and time expressions.
• Improved German and Spanish AI engines with fewer odd or incorrect suggestions.

Enhancements

Spelling

• Significant improvements across all languages with major English, German and Spanish dictionary expansion.
• New common word database for all languages with 461 proper names and 212 abbreviations to be recognized as valid cases.
• Improved the case sensitivity approach in suggestions and exclusions from suggestions for added dialectal cases, affixes, technology, common, and proper name terms, chemical elements, shortened abbreviations, common, abbreviations, proper names, archaic, slang, and technology terms.

Autocorrect

• Removed invalid autocorrect pairs where archaic or dialectal pattern is a valid word.
• Added new prioritization pairs for proper name misspellings and common typos or misspellings.

Style

• Validated and extended dialectical, doubtful, barbarisms, profane patterns for English, German, Ukrainian, and Swedish. 
• Added style guide entries for LGBTQIA+ and disability-inclusive language for English.
• Significant Ukrainian style guide enhancement (1,624 entries) for anglicisms, antysurzyhk, and profanity categories.

Grammar

• Third-party grammar rule engine improvements across German and Spanish with a focus on reducing incorrect punctuation suggestions.
• Improved the mechanism of case sensitivity handling and compound word processing.

AI proofreading 

• Release of a new version of an in-house AI-driven proofreading engine for English which fixes bugs with en dashes/em dashes/hyphens, pluralizations, medical terms, split sentences, past and future time after 2021, cases with time patterns am/pm/AM/PM to a.m./p.m.
• Improved proofreading engines for German and Spanish, in particular, for nonexistent, incorrect spelling, incorrect punctuation, and weird suggestions.

📝 TL;DR

WProofreader SDK v6.3.0 delivers:

• New disableTypeInSuggestionPopup option and onUsage callback.
• TinyMCE 8 support for the WProofreader plugin.
• Improved custom dictionary exclusions and English suggestion ranking.
• Rolled back AI Spanish engine default config to prevent overcorrection of medical terms.
• Fixes for British English autocomplete (on-prem) and English autocomplete outage (cloud).
• Security updates to system prompts for multiple languages in AI writing assistant.

🛠️ Enhancements

✨ AI writing assistant

• Updated system prompts to improve prompt injection protection for Dutch, Swedish, German, and French.

WProofreader JS core 3.30.5916

• Added a new boolean option disableTypeInSuggestionPopup to hide suggestion type labels in the suggestion pop-up. Disabled by default.
• Introduced a new onUsage callback function, triggered whenever a check, autocorrect, autocomplete, or generate request is executed and tracked. Returns key metadata about the request type along with selected input and output values. Useful for monitoring usage, collecting metrics, and supporting client-side billing or analytics.
• Updated description of the onStatistics callback function. Triggered when a user performs a suggestion or dictionary action, returning details such as action type, input text, rule, language, and context.

More details about onUsage and onStatistics are available in the WProofreader configuration reference.

• Fixed an issue with the user-level dictionary where, in some cases, words could be re-flagged even after being added. This occurred since v5.35.1 due to the introduction of rule IDs for spelling suggestions.
  
• Added support for TinyMCE 8 in the WProofreader plugin (now supports TinyMCE versions 6–8). Try the demo.

📚 Spelling

• Improved exclude custom dictionary functionality. Added a case-sensitive flag to allow excluding a word only in its exact form, keeping other forms correct. Without the flag, all forms of the word (including possessives) will be excluded. Available in the on-premises (self-hosted) version. This applies both to excluding words from being correct and from appearing in suggestions (while remaining correct).
• Improved prioritization of English spelling suggestions generated by the algorithmic engine using n-gram frequency data. Added a new parameter NgramPrioritization in AppServerX.xml.

AI proofreading

• Rolled back the default configuration for the AI Spanish engine due to an increase in incorrect/ignored suggestion reports. The AI model was removing unknown words, particularly medical terms.

💡 Autocomplete

• [On-prem] Fixed a critical issue causing British English autocomplete suggestions to be non-functional since v6.1.0 due to internal misconfiguration.
• [Cloud] Fixed an issue where English autocomplete suggestions were unavailable from July 25 to August 1, 2025. The issue went unnoticed as the system returned 2xx responses with empty results.

🐞 Bug fixes

• [WP-6507] WProofreader core: Broken CSS styles in Webflow site builder
• [WP-5345] Spelling: Misspelled word “gioa” in Italian isn’t flagged

📝 TL;DR

Expanded Ukrainian, English, Spanish, and Portuguese dictionaries, added Ukrainian AI assistant UI localization, and patched CVE-2024-45772.

🛠️ Enhancements

📚 Linguistic updates

• Major Ukrainian dictionary expansion with 2,145 new words including anglicisms, slang, loan translated words and legal terminology. English receives 207 new words across multiple domains with significant medical term cleanup (191 removals), Spanish — up-to-date medical terminology, Portuguese — 144 new split patterns for prefixes and suffixes.

🛡️ Security

• Fixed vulnerability CVE-2024-45772 in lucene-core.jar (grammar engine, third-party dependency) by upgrading to version 9.12.0, which includes the official security fix.

✨ AI writing assistant

• Added UI localization for Ukrainian.

📝 TL;DR

• Claude 3.5 and Mistral AI models replace deprecated Claude Instant 1.2
• AI writing assistant now supports Ukrainian
• Style guide adds context awareness and language-specific rules
• Docker environment variables now use “WPR_” prefix
• Security fix for CVE-2025-48924
• Linguistic updates across English, German, Spanish, Ukrainian, and French

🛠️ Enhancements

✨ AI writing assistant

• Added support for Claude 3.5 Sonnet v1 and Mistral AI models.
• Enabled Ukrainian language support.
• Improved prompt localization and wording for English, French, Spanish, and Portuguese to prevent prompt injection and jailbreak attacks.
• Introduced MaxGenerateInputSize and MinGenerateInputSize parameters in AppServerX.xml configuration for AI input control.
• Increased Bedrock InvokeModel call timeout to improve request stability for languages like Ukrainian.

⚠️ Important change

• Prior to v6.2.0, the AI writing assistant used Claude Instant 1.2 as the base model. This model was deprecated by Amazon and reached its end of life on July 21. Due to the lack of advance notice, support was urgently replaced with Claude 3.5 Sonnet v1 and Mistral AI. To continue using the AI assistant for non-English languages, it is required to upgrade to v6.2.0.

Style guide

• Implemented context awareness to better match real-world usage.
  Added context_include and context_exclude fields to define additional words or phrases that trigger rule activation. Available via API or in the style_guide.csv file (for the self-hosted version).
• Allowed creation of language-specific rules in the style guide API.
  In addition to common (language-independent) rules, it’s now possible to create language-specific rules using the lang parameter. Available via API or by placing style_guide.csv under the respective language directory (for the self-hosted version).
• Added support for ignoring hidden files in the style guide directory.

WProofreader JS core v3.29.5897

• Removed ‘Show more’ button when only one suggestion is available.
• Improved autocorrect handling to prevent overlapping suggestions and allow revert actions. Highlighting is now delayed in the same position until the user types three or more characters or reverts the change, reducing UI conflicts and improving interaction flow.

Spelling engine

• Restricted split suggestions and refined rule application logic.

📚 Linguistic updates

• Significant linguistic improvements across five languages with major Ukrainian dictionary expansion (267 new words) and comprehensive English updates (334 additions, 285 medical term removals). German receives 75 new words and enhanced style guides, while Spanish gets improved autocorrect filtering and grammar rule updates. French introduces anglicisms and strong language detection with 187 new style guide entries.

⚠️ Breaking changes

• Docker deployments: All environment variables and build arguments now use the WPR_ prefix. Check the updated Dockerfiles and README.

🐞 Bug fixes

• [WP-6411] Spelling: Restrict application of WORD_INITIAL/WORD_FINAL if there is hyphen
• [WP-6412] Spelling/prioritization: Restrict partial-match rules from applying to full-word matches
• [WP-6454] Spelling engine: Allow hyphen at the end if the original word includes a hyphen
• [WP-6478] WProofreader core: Dialog shows “Well done…” even when turned off
• [WP-6449] WProofreader core/autocorrect: Can’t revert autocorrect when additional suggestions are applied to the same word
• [WP-6479] WProofreader core: Broken styles for badge and its menu
• [WP-6485] Linux installer: File permissions aren’t set correctly for service files and models
• [WP-6494] Security: Fixed CVE-2025-48924 in org.apache.commons:commons-lang3

📝 TL;DR

Significant linguistic updates across English, German, and Spanish spelling engines, including new domain-specific terms, improved suggestion prioritization, and cleanup of invalid entries. English spelling improvements include 172 new FDA medical terms and 153 refined misspelling patterns. Grammar engine refinements address 24 incorrect suggestions, and autocomplete filtering has been expanded to better handle sensitive content.

🚀 Enhancements

📚 Spelling engine

🌐 English

• Added 620 words across domains including abbreviations, biology, business, common usage, dialectal terms, Latin, linguistics, measurements, medical, neologisms, proper names, science, slang, sport, and technology. Removed 21 words from dialectal, common, archaic, and linguistic categories.
• Added 172 medical terms based on FDA-approved drug names from 2024 and 2025. Removed 281 medical terms due to issues such as non-standard capitalization, Latin forms, and non-existent terms.
• Excluded 35 words from suggestions, including certain abbreviations, medical terms, Greek-origin words, and proper names.
• Improved prioritization of 153 common misspellings and phonetic variants, promoting more accurate suggestions to the top of the list.
• Improved dialect handling by removing 19 word pairs that were invalid, misspelled, or not dialect-specific.
• Updated split patterns for run-on words: added 4 patterns (prepositions, prefixes, suffixes) and removed 1 preposition-related pattern.
• Removed 92 invalid autocorrect entries, including incorrect patterns and suggestions.
• Removed 1 invalid suggestion previously returned by the AI GEC engine for a non-existent word.

🌐 German

• Added 146 words related to proper names (e.g., names of well-known companies such as Apple), common usage, slang, technology, culinary, and anglicisms.
• Added 25 medical terms.
• Improved prioritization of 98 entries, including capitalization of anglicisms and correct splits, placing more relevant suggestions at the top of the list.
• Removed 4 entries with incorrect spelling autocorrect suggestions.

🌐 Spanish

• Added 825 words related to administration, biology, chemical elements, common usage, ecology, economics, linguistics, math/logic, medical, politics, proper names, psychology, science, sport, and technology. Removed 3 incorrect words.
• Added 99 medical terms.
• Removed 6 spelling autocorrect entries due to incorrect splits or malformed word forms.

⚙️ Grammar engine

🌐 English

• Disabled 23 incorrect suggestions generated by the AI GEC engine.
• Disabled 1 rule from LanguageTool where “Check box” could be incorrectly flagged as separable.

💡Autocomplete suggestions

🌐 English

• Added 20 terms related to violence and mortality to improve filtering of completions shown to end users when text prediction is enabled.

Enhancements

Grammar engine

• Updated the third-party grammar engine, LanguageTool, to version 6.6. This version includes added and improved grammar rules for Catalan, English, German, Portuguese, Spanish, and Ukrainian.
• Updated the vulnerable logback-core.jar library used in LanguageTool v6.6 to address the following security issues: CVE-2024-12798,  CVE-2024-12801

Important: LanguageTool v6.6 now requires Java 17 (previously Java 8).

Enhancements

• WProofreader core (version 3.20.5729)
  • Proofreading dialog in sidebar. Introduced a new option, proofreadDialogContainer, enabling the Proofreading dialog to be displayed in a reserved section of the page, such as a sidebar. This option integrates the proofreading experience directly into the page layout. 
  • Faster error detection and highlighting while typing. Introduced a new option, checkOnSpace, to improve the responsiveness of error highlighting as you type. This option triggers additional checks whenever a space is typed, resulting in faster detection and display of issues. However, enabling this feature increases the number of backend requests and the amount of text processed (word count). By default, it is disabled. For more details about these new options, see the configuration reference. 

• Spelling engine
  • Enabled the use of suggestions from the spelling cache for requests that include custom dictionaries (both user-level and organization-level).
  • Added spelling autocorrect files with pre-validated misspelling-suggestion pairs for the New Zealand (en_NZ) and South African (en_ZA) English language variants.
  • Updated English dialect-specific spelling validation functionality (covering American, British, Canadian, and Australian English). Added 80+ medical terms with ‘hemo-/haemo-‘ prefix variations and removed incorrect entries (affecting variations like ‘jeopardize/jeopardise’, ‘tokenize/tokenise’, and medical terms with incorrect dialect mappings).
  • Updated core English lexicons. Added 300+ terms including technical abbreviations, computing terminology, cyber-related vocabulary, and common abbreviations (e.g., Ph.D., Inc.) and Latin terms (e.g., summa laude, terra firma). Also, restructured word lists: created dialect-specific files for allowed and excluded words.
  • Updated English medical dictionaries. Added 160+ clinical terms (biopharmaceuticals, medical device terminology) and removed 320+ entries (non-medical terms, outdated clinical spellings, and incorrectly formatted medical abbreviations); minor updates to legal dictionary.
  • Enhanced spelling autocorrect functionality. Added approximately 170 new corrections per dialect, including new support for South African and New Zealand English. 
  • Updated list of excluded suggestions, added cyber-related terms, removed term with neutral meaning.

• Grammar engine
  • Updated 3rd party grammar engine, LanguageTool, to version 6.5.0. Added and improved grammar rules for the next languages: Catalan, Dutch, English, French, German, Portuguese, Spanish and Ukrainian.
  • Updated vulnerable ‘commons-io’ library used in third-party grammar engine, LanguageTool v6.5.
  • Disabled unnecessary corrections and specific rules in both rule-based grammar and AI GEC engines to improve handling of various writing contexts (casual expressions, technical terms, sports terminology, and business language).
• Style guide
  • Enhanced normalization of dashes and hyphens to ensure custom rules recognize and match variations of these characters.
  • Fixed incorrect mapping of letter cases in style guide suggestions, particularly for mixed-case scenarios like Tall Man Lettering (TLM).

• Autocomplete “text completion” engine. Filtered invalid text completions containing numbers.

• Check API
  • Added a new query parameter, exclude_wordlist, for the check and check_spelling API commands. This parameter accepts a comma-separated list of words to be considered incorrect during checks. It works similarly to user_wordlist but excludes the specified words. The parameter is case-sensitive and language-agnostic.
  • Added a new query parameter, check_kit, for the check API command. This parameter enables control over the text checking mode by selecting specific engine combinations (AI-powered, algorithmic-only, or mixed). If not specified, the default configuration will be applied. For languages where the specified configuration is not available, the request will fail with an error message.

Bug fixes

• [WP-5942] AI GEC engine: Dashes treated as part of the word
• [WP-5958] AI GEC engine: Glued suggestion when replacing dash with ‘and’
• [WP-5968] AI GEC engine: Glued words in suggestion when removing dash
• [WP-5955] API: Misleading error message “Failed to decode URI.”
• [WP-5967] WProofreader core: Incorrect timeout (2s) for check request in single paragraph mode
• [WP-5927] Spelling engine: Duplicated suggestions for dialect-specific issues

Enhancements

• Style guide
  • Added a new collection of LGBTQIA+-inclusive rules for all English dialects. The updated rules will suggest gender-neutral language, such as “her spouse” or “his spouse” instead of “her wife” or “his husband”. The rule category name is WSC_LGBTQIA_INCLUSIVE_LANGUAGE.
  • Added a new action “upload” to the style guide API that allows uploading CSV files with multiple rules.
• [On-premise] Application server. Increased default number of worker threads to 16 to improve performance.

Other

• Demo pages. Enhanced WProofreader SDK demo pages with a quick demo tour to show the main aspects of the WProofreader interface and how to effectively use the demo.
• [Cloud] Grammar engine. Rollback of the 3rd grammar engine, LanguageTool v6.4 to v6.3 due to potential issues with CPU utilization under load.

Bug fixes

• [WP-5725] AI engine-only configuration for English doesn’t catch dialect differences
• [WP-5771] AppServer: “enforce_ai” parameter overrides default check kit [On-premise]

Enhancements

• WProofreader core v3.15.5524
  • Added an experimental “checkInNonEditableElements” option that enables text checking in non-editable elements. It can only be used with the ‘init’ and ‘autoCreate’ initialization methods (it doesn’t work with ‘autoSearch’). Disabled by default. More information and other options are available here.
  • Disabled checking in code and formula blocks of Quill WYSIWYG editor.
  • [On-premise] Removed “Report incorrect” action if the database isn’t configured.
  • AI writing assistant. Improved text formatting in suggested completions by adding new lines for readability.
  • Updated WProofreader SDK JavaScript NPM package with basic TypeScript support. The package includes the type definition file (.d.ts) for the WProofreader core methods and additional NPM package methods, covering initialization, configuration, and instance management.
• Spelling engine. Extended English spelling dictionaries with 21 correct words previously unrecognized by the spelling engine and terms reported by clients (e.g., redshift, nuclease).
• Other
  • WProofreader Helm Chart. Updated to version 1.0.0, including enhanced README documentation on recommended resources, limits, and scaling for Kubernetes pods. Learn more about the updates.

Bug fixes

• [WP-5741] Spelling: Word “testin” isn’t recognized as a misspelling in American English (en_US) and Canadian English (en_CA)
• [WP-5662] WProofreader core / AI writing assistant. User text selection is lost after opening assistant dialog [Gmail]

Enhancements

[Cloud] The initial release of Triton Server enhances efficient deployment and scalable inference of AI language models. This update is exclusive to the cloud version, with no changes made to the on-premise package.

Enhancements

• Named entity recognition (NER). Implemented a new functionality for seamless identification of entities such as names, locations and organizations. This functionality ensures that named entities are correctly identified and not flagged as misspellings.
  • Available for English, Spanish, German languages and their dialects.
  • Enabled by default for English, while for Spanish and German, it’s in a disabled state.
  • [On-premise] In the on-premise package, the NER functionality is added as a separate NERModel tag with “Enabled”, “Path”, ”ProbabilityThreshold” settings inside the respective language configuration of the AppServerX.xml file. Using “Enabled” attribute the feature can be enabled or disabled.
  • [On-premise] Adding NER affects minimal hardware requirements. If any of the English, Spanish, German languages is chosen to install, an appropriate NER model will be automatically downloaded. Each NER model requires around ~350 MB.
• Spelling & grammar engine
  • The AI-powered engine is enabled for the following language configurations:
    • American English (en_US) and New Zealand English (en_NZ). The AI-powered configuration is enabled by default for all clients using the cloud service. It can’t be disabled in WProofreader SDK using the “enforceAI:false” parameter.  
    • Mexican Spanish (es_MX). The AI-powered configuration isn’t enabled by default but can be turned on by using the “enforceAI:true” parameter or “enforce_ai=true” for API requests.
    • [On-premise] If Spanish, German and/or English AI language models are selected to install, the AI-powered engine will be added to the language configuration along with the algorithmic spelling and grammar engines. But this can be updated manually in the AppServerX.xml file. 
  • 3rd party grammar engine. Updated vulnerable versions of Jetty/Netty libraries shipped with LanguageTool package. 

Bug fixes

• [WP-5163] WProofreader core: Misleading error message when language code is not available
• [WP-5230] WProofreader core/Google Docs: Highlighted error are merged into one and are not divided into parts
• [WP-5152] AI German: Post-processing adds excessive spaces in suggestions with quotes
• [WP-5242] AI English: Post-processing provides broken suggestions for numbers with commas
• [WP-5166] [On-premise] WProofreader SDK demos: Error when working with enforceAI if model isn’t installed
• [WP-5195] [On-premise] Web server is not configured properly after installation
• [WP-5193] [On-premise] Linux installer/Docker: Abort build if language files were not downloaded
• [WP-5245] Spelling autocorrect: 400 error when American English (en_US) language is not enabled
• [WP-5253] Autocomplete parameter is set to “true” if respective model is not installed
• [WP-5258] [On-premise] AppServerX.xml: Incorrect left-to-right (LTR) mode is used for Urdu (ur_PK) language
• [WP-5228] English medical dictionaries: Incorrect word ‘Helicobactor’ is present in dictionaries