Here at Yaraku, we believe generative AI and LLMs can contribute significantly to productivity and efficiency.
New models with better intelligence and more capabilities are being developed all the time. And to take advantage of these improvements, we will be making some updates to the current models that are available in Yaraku Translate.

There will be no downtime due to this update.

Change date
02/04 (Wed),（JST）

1. GPT 5-mini / GPT 5.1

Replacing our current default engine GPT 4.1-mini will be GPT 5-mini. Our GPT 4.1 will also be replaced with GPT 5.1
The GPT 4.1-mini model had faster processing times, allowing us to increase the character limit for that engine to 150,000 characters. GPT 5-mini will continue in that vein with fast processing speed and high translation quality.

GPT 5.1 is the next step in the GPT 5 series. From OpenAI’s website:
“It now adapts its thinking time more precisely to the question—spending more time on complex problems while responding more quickly to simpler ones.”

For those concerned about security, please rest assured. Our Business Associate Agreement still applies so the same security and data protections will continue to be available with the new model.

2. Claude 4.5 Sonnet

Developed by Anthropic, Claude is another option alongside GPT.
Our Claude 3.5 Sonnet will be updated to Claude 4.5 Sonnet. Some of our users have also found that Claude produces more natural-sounding translations as compared to other generative-AI options. If that is required for the documents you work on, give Claude a try.

The newer model also has a bit more capacity, allowing us to raise the character limit for Claude from 50,000 characters to 70,000 characters.

3. Gemini 3.0 Flash

Part of Google’s Gemini 3 era, Gemini 3.0 Flash is the speed optimized version of their Gemini 3.0 model.

From Google’s website:
“With Gemini 3, we introduced frontier performance across complex reasoning, multimodal and vision understanding and agentic and vibe coding tasks. Gemini 3 Flash retains this foundation, combining Gemini 3’s Pro-grade reasoning with Flash-level latency, efficiency and cost.”

4. Behind the Scenes

We sometimes get asked why we choose certain models, or why we don’t update to the latest models. In fact, behind the scenes a lot of work is done before we make a decision.

Our in-house NLP team does a comprehensive evaluation of new models and compares the results to the performance of currently adopted models. Their evaluation takes into account not just translation quality (BLEU and BERT score), but also the speed of the response, and the cost per character. Then, we try to find the best compromise between these 3 factors.

And unfortunately, sometimes new is not always better. Past evaluations have shown that sometimes a newer model that is more advanced might not be so good at translation due to the type of intelligence. Or, it might take too long to respond which would impact the speed of the translation process. In some rarer cases, we also find models that are worse in translating specific languages.

Only with all these data points in hand do we make a choice on the best model for our application and for our customers.

Thank you.