Machine Translation Just Got a Lot Bigger
This is a big one for reach. Two connector upgrades, shipped in a point release days after the main one, expand your machine translation coverage in a serious way. The DeepL connector now requests DeepL's next-generation models for every language pair, with automatic fallback to the classic model where the new one is not ready yet. Those models raise quality on existing pairs and add 89 newly supported languages, including regional variants like Chinese Simplified and Traditional and Latin American Spanish. Existing DeepL connectors get the upgrade automatically, and you turn on the new languages from the connector's language list. A few of the newest ones do not support glossaries or inline tags yet, so they return as plain text, while every pair you already use keeps working. The specifics are in the DeepL connector guide.
The Microsoft Translator connector adds 31 more languages and fixes several outdated language codes that had been causing silent failures. Existing setups keep working, and obsolete selections clean themselves up next time you save. Between the two, that is well over 100 new language options without touching your configuration. For teams automating across clients, the MT Hive API can now run a request using a specific client company's profile, loading that company's engine, memories, parsers, and formats while the request stays yours.
Webhooks That Look After Themselves
If you build on the platform, webhook reliability got real love this release. An endpoint that keeps failing is now disabled automatically: return HTTP 410, or fail ten deliveries in a row, and the webhook switches off and notifies your admins, with the reason shown right on the webhook page. Re-enable it or change the URL and the counter resets. On top of that, deliveries to slow endpoints that blew past the 30-second window used to be misread and dropped silently, which broke monitoring and fired false alarms. Those timeouts are now marked correctly and retried like anything else. It is all in the webhooks documentation, part of the wider API toolkit that developers live in. A new API endpoint also lets authorized users delete a single import batch from a translation memory or term base, running asynchronously with an operation ID to track.
One File Filter for Any Language Pair
The multilingual Excel parser gains two placeholders, Any Source Language and Any Target Language, that resolve to a project's real locales at parsing time. Instead of maintaining one configuration per language combination, you build a single filter that works across any project, a real time-saver for anyone managing dozens of language pairs. When Any Target Language is used, each job produces its own bilingual deliverable with the source preserved and the target filled by the translator. Existing setups are untouched, and the steps are in the Excel parser guide. The CSV parser picks up the same monolingual-by-column and word-level do-not-translate options Excel already had, and file upload and preparation now run as one flow, with a checkbox that carries you straight into the preparation wizard.
Fixes That Protect the Document
Six fixes guard content that used to slip away quietly. Whitespace between consecutive self-closing inline tags in XML with HTML content is no longer dropped on export. Single non-breaking and soft hyphens in Word documents now show as visible markup and survive into the translated file. DOCX exports with Chinese, Japanese, or Korean text no longer corrupt, because the font filter uses an explicit allowlist of safe fonts. Color-based do-not-translate text at the start or end of a cell or paragraph is kept as a hidden tag, so translators can see and reposition it for different word orders. Translation memory entries with no target translation no longer come back as 100% matches in XLIFF API responses, so nobody trusts an empty match. And interpreting orders submitted through the New Order Form once again create the interpreting job automatically, with the right appointment times, instead of being skipped.
Also in This Release
A few more refinements round it out. The New Order Form can accept up to 250 files per order, up from 60, so big batches skip the zip workaround. Two opt-in columns, Order Reference and Order ID, are now in the Standard jobs list, with the project-list Order ID gated by order access rights. Automatic job slicing now spreads paragraphs evenly across jobs and lets you cap the number of jobs per document, which makes workloads and cost allocation fairer; the details are in the job slicing guide. The Supplier Quotes and Invoices dashboard now shows Job IDs for Codyt jobs, matching Standard, a small win for finance reconciliation. The Markup QA rule gains an optional Check Tag Order setting, so translators can reorder tags for different word orders while Wordbee still verifies they are present and valid. And the Google Cloud Translation connector points to current docs with accurate field descriptions.
A release that stretches how far and how reliably you can translate. The full technical notes for 2026.2.7.0 live on help.wordbee.com. If broader machine translation or self-healing webhooks change what you can automate, talk to us and we will help you put them to work.