MachineTranslation.com has announced it has expanded the AI model pool that powers its SMART consensus mechanism. With the addition of Aya Expanse 32B by Cohere and MiniMax M2.7, the platform now compares the outputs of up to 24 leading AI models, including ChatGPT, Claude, Gemini, DeepL, and Google. SMART surfaces the translation the majority of them agree on as the result every user receives.

The expansion is more than a feature update. Every model added to the pool means another independent voice cross-checking every translation before it reaches a user. Internal benchmarking shows this consensus approach reduces critical translation errors and hallucinations by up to 90% compared to any single-model baseline.

What each new model brings

Aya Expanse 32B (Cohere) is a multilingual model built for 23+ languages, with particular strength in low-resource and non-English language pairs where single-model tools are most likely to fail silently.

MiniMax M2.7 is a high-performance model recognized for reasoning depth across Asian languages, including Chinese, Japanese, and Korean. These are language pairs that demand more than surface-level fluency.

“Every time we add a new model to MachineTranslation.com, we’re not just expanding a list. We’re increasing the quality of the translation,” said Ofer Tirosh, CEO of Tomedes.

“Aya Expanse 32B fills a gap we’ve been tracking carefully: underserved multilingual markets where a single model’s cultural and linguistic blind spots go unchecked. MiniMax M2.7 brings serious reasoning depth to Asian language pairs. Together, they don’t just add coverage. They make the translation that the majority agrees on even harder to get wrong.”

Rachelle Garcia, AI Lead at Tomedes, added: “The more linguistically diverse the model pool, the harder it is for any one model’s bias to go unchallenged. Aya Expanse 32B was built to close the quality gap in language pairs where single-model tools fail most. When it joins the SMART pool, it actively protects the translations where protection is needed most. Our data shows that 82% of translation errors on professional content occur in exactly these underserved language corridors.”

How SMART works

Unlike every other AI translation tool, MachineTranslation.com does not ask users to trust one model’s answer. Through SMART, the platform makes its full pool of leading AI models produce independent translations of the same text simultaneously. The translation the majority agrees on, is the one users receive. Every output is AI-verified before delivery.

Frequently Asked Questions

• Can users see how different AI models translated their text? Yes. MachineTranslation.com displays each model’s output individually alongside the agreed-on result, so users can see exactly where the AI systems aligned and where they didn’t.
• Why does MachineTranslation.com use multiple AI models instead of just one? A single AI model has no way to check if it is wrong. MachineTranslation.com was built on the principle that the translation dozens of independent AI systems agree on is fundamentally more reliable than the translation any one of them produces alone.
• Why Aya Expanse 32B and MiniMax M2.7? Both were selected for demonstrated strength in the language pairs where single-model tools are statistically most likely to fail: low-resource multilingual languages and Asian language pairs respectively.
• Is Human Verification still available? Yes. For any translation requiring 100% guaranteed accuracy, Human Verification is available as an add-on within the same platform, with no external agency required.

About MachineTranslation.com

Developed by Tomedes, MachineTranslation.com is a leading AI translation tool. By using SMART to compare up to 24 AI models and select the translation the majority agrees on, the tool provides secure, transparent, and highly accurate translations for over 1.5 million users worldwide.