ModelFront Announces General Availability of Automatic Post-Editing

Additional LLM to grow automation now included by default for all customers.

PALO ALTO, Calif.–(BUSINESS WIRE)–ModelFront today announced the general availability of automatic post-editing (APE), an additional private custom large language model.

First rolled out in 2024, APE is now rolled out to all ModelFront customers in production and included by default from day one.

ModelFront’s quality prediction (QP) is AI to check and fix AI translations. It automatically verifies segments that humans would verify unedited, and triggers human intervention for those that need it. Enterprise translation buyers use QP to automate and scale while keeping human quality.

While QP alone successfully automated segments that professional human translators were verifying totally unedited, many segments still sent to humans required only repetitive, mechanical edits. APE generates these edits, to grow what is automated, while still keeping human quality. Human intervention is still triggered for those segments that require human intelligence, research or decisions.

“This was a logical next step, that customers pushed for,” said Adam Bittlingmayer, CEO and technical co-founder of ModelFront.

“This was a logical next step, that customers pushed for. In theory, these are fixes that could be made by custom machine translation models, like Google or Microsoft.

But in reality, it never happened. Enterprise translation teams can’t train thousands of models to cover all combinations of language, content type and workflow step. ModelFront models are built to learn the workflows and built to work together.”

– Adam Bittlingmayer, CEO, ModelFront

APE was first deployed in production in 2024 and was successfully rolled out across ModelFront’s customer base — primarily Fortune 500 companies and their content — by the start of 2026.

Advanced buyer teams shared results publicly at top industry events, like LocWorld 2024 in Monterey. To date, ModelFront APE has resulted in hundreds of millions of additional automated words for enterprise translation buyers.

“You can get a very substantial increase in your MT autoapproval rates without sacrificing quality” said Conchita Laguardia, the specialist running AI and Tech for Localisation inside Farfetch (Coupang (NYSE: CPNG)), one of the first enterprises to adopt quality prediction and automatic post-editing.

“Ever since I implemented MTQE in production and saw it worked, I thought that automatic post-editing was the natural evolution step after it.

If the model can detect a bad translation, surely it can also attempt a fix and re-evaluate the fixed output?

While it won’t fix everything, APE makes sure that systematic errors are no more, by embedding those recurring human fixes into a closed-loop optimization cycle that first detects, then corrects, then self-evaluates the correction.

Depending on the language pair, you can get a very substantial increase in your MT autoapproval rates without sacrificing quality.”

– Conchita Laguardia, AI and Tech, Localisation, Farfetch

ModelFront APE is available across all ModelFront integrations, including Phrase (The Carlyle Group (NASDAQ: CG)), XTM (K1 Investment Management), memoQ, WorldServer and Trados Enterprise (RWS (LSE: RWS)) and GlobalLink and Wordbee (TransPerfect), and via the ModelFront API.

ModelFront APE works with AI-generated translation from any source, including Google Translate and Gemini (Alphabet (NASDAQ: GOOGL)), DeepL, Microsoft Custom Translator (Microsoft (NASDAQ: MSFT)), OpenAI, Claude (Anthropic) and Systran (ChapsVision).

As part of ModelFront’s system, APE is an additional private custom LLM built with the same strict data privacy guarantees. Customer data is never sent to generic shared LLMs or any third-party AI.

With ModelFront, APE always works together with quality prediction (QP), a separate LLM that maintains human quality. QP verifies translations, APE generates edits. Edits generated by APE are always resent to QP for verification, just like the original untrusted raw AI translations are. APE that QP cannot verify is sent to professional human translators. APE only creates significant value together with QP.

ModelFront APE is used by translation teams inside large enterprises across industries like software, law, fashion, travel and pharma.

Availability: Automatic post-editing is now included in ModelFront by default for all customers.

About ModelFront

ModelFront is AI to check and fix AI translations and trigger human intervention as needed, to scale translation while keeping human quality.

GenAI started in translation, but failed to accelerate human-quality translation. Manually checking every AI translation (“post-editing”) was still too slow and expensive.

Companies use ModelFront to fully automate millions of words, while keeping the same human quality, right inside their existing setup.

Unlike most AI providers, ModelFront provides verification (✓ or ✗), not just unverified generation, and take responsibility for keeping human quality.

Hundreds of millions of words of high-value Fortune 500 content have been trusted to ModelFront.

ModelFront does not provide manual human translation services. ModelFront is a Gartner Cool Vendor and recognized by translation industry publications like Nimdzi, Multilingual Magazine, Slator, CSA Research. ModelFront Inc. is based in Palo Alto, California.

ModelFront’s vision is more content in more languages for more people.

About translation

Generative AI started in translation – the Transformer model architecture, the foundation of large language models, was invented for machine translation at Google in 2017.

Translation was thus the first workflow to shift to humans verifying and editing AI output, rather than manually writing content from scratch.

Translation is also key to human progress. More than 6 billion people cannot understand English, and companies spend $70 billion a year, and growing, on translation and localization.

Contacts

Media Contact:
ModelFront Inc.
press@modelfront.com
https://modelfront.com