Gemini draws on personal data as Google introduces TranslateGemma

Google introduced Personal Intelligence as a beta in the Gemini app on January 14, 2026.¹ With permission, Gemini can use information from Gmail, Google Photos, Search history, and YouTube history to tailor its answers.¹ Rollout began for Google AI Pro and AI Ultra subscribers in the US and is limited to personal Google accounts.¹

The feature is off by default and can be enabled in Gemini settings.¹ It draws on Google's own services to give the assistant more personal context as AI companies compete to make assistants more useful to individuals.¹² Workspace business and education accounts cannot use it at launch.¹ Google plans to expand it to more countries and the free tier but has not given a date.¹

The Wikimedia Foundation publicly named more customers and partners of its paid Wikimedia Enterprise service on January 15, 2026.¹ They include Amazon, Meta, Microsoft, Mistral AI, and Perplexity.¹ The agreements were made over the preceding year; the announcement was not the launch of a new paid service.¹² Enterprise APIs give large commercial users structured access to Wikimedia content as an alternative to retrieving pages at scale.¹

For AI service providers, this offers a contractual route to content they can use in their products.¹ Wikimedia Enterprise existed before these agreements.³¹ Wikimedia says the revenue supports the Wikimedia movement.¹ It disclosed no prices or contract values.¹² The announcement also does not detail how each named company uses the content in its AI products.¹

Google introduced TranslateGemma on January 15, 2026, as an open family of translation models based on Gemma 3.¹ The family includes 4B, 12B, and 27B parameter models designed for hardware ranging from phones to cloud accelerators.¹ The models are available through Hugging Face, Kaggle, and Vertex AI.¹²

Google positions the smallest model for phones and edge devices, the 12B model for laptops, and the 27B model for more powerful hardware.¹ In Google's translation tests, the new 12B model outperformed an earlier 27B baseline.¹ That gives companies options for running translation with different levels of computing capacity, including on-device deployment. Before choosing a model for production, they should test its performance on their own language pairs and specialized texts rather than rely solely on the published results.

Black Forest Labs introduced the FLUX.2 [klein] model family on January 15, 2026. Its architecture supports text-to-image generation, image editing, and workflows using multiple reference images.¹ The models are available through an API, for local use, and in a demo app.² The family includes variants with 4 billion and 9 billion parameters.¹

The focus is on reducing wait times for image creation and editing. Black Forest Labs says image generation can take less than a second, though that should not be treated as a performance guarantee for every use case.¹² Cloudflare also offers the smaller variant through Workers AI.³ Licensing is an important distinction for businesses: the 4B variant uses Apache 2.0, while the 9B variant’s open weights carry a noncommercial license.¹

OpenAI has offered a separate translation page at chatgpt.com/translate since mid-January 2026.¹² Its two-pane interface accepts text, detects the source language automatically, and translates into a selected target language.³⁴ Suggestions for changing the style, such as business-oriented or simpler wording, take users into ChatGPT for further editing.³⁵

The page provides a dedicated interface for a task that was already possible in ChatGPT.¹⁵ It competes with Google Translate and DeepL but does not cover all their functions: document and website translation and image features were not consistently available at launch.¹⁴ OpenAI has not disclosed which model powers the page, and the interface alone offers no basis for judging translation quality.¹²

Anduril is continuing tests of Fury, its autonomous fighter aircraft designated YFQ-44A by the US Air Force.¹² By mid-January, the company said it had multiple completed aircraft in testing and was integrating weapons.² On January 22, a company spokesperson said its Long Beach site would focus on research, development, and work adjacent to manufacturing, while higher-volume manufacturing would take place in Ohio.³

Fury is intended to fly as an autonomous or semi-autonomous wingman alongside crewed fighters such as the F-35.¹² The next steps involve more than flight testing: weapons integration and the transition to manufacturing are still ahead.³² Anduril has not given a firm start date for higher-volume manufacturing or a unit price.³²