<![CDATA[Background: The rapid integration of AI in machine translation has spurred extensive research, with over 2,000 studies (2024–2026) on translation quality, mostly focused on general and literary texts. This leaves a gap in evaluating complex political texts, where cultural nuances and specialized terminology pose unique challenges. Purpose: This research is carried out with the objective to examine level of accuracy, acceptability, and readability of the translated results of French political news texts translated by DeepSeek and Gemini using Karangturi framework belongs to Honggo to find a model of AI that shows excellent translation quality based on Nababan's accuracy and readability theory and Machali's acceptability theory. Method: The research method used is descriptive qualitative with research data taken from three news portals in France, namely Le Figaro, Libération, and Le Monde, each of which represents different political ideologies. The research instrument uses Nababan's accuracy and readability theory and Machali's acceptability theory, modified by Ardi. Results and Discussion: The test results show the same accuracy score (87.04%) and acceptability (82.64%) by both AI machines. However, Gemini is more excellent in readability with a score of 98.61% compared to DeepSeek, which scored 93.06%. Therefore, in total, Gemini (89.65%) better than DeepSeek (87.63%). Conclusions and Implications: This study concludes that Gemini and DeepSeek have great potential in translating political news texts, but their translation results still require human editing due to limitations in grammatical and contextual aspects. These results can be used as an academic reference for researchers and students to be more critical of the role of AI in the transfer of socio-political meaning, and also suggest that future researchers explore other AI frameworks or machines to expand the scope of this research.]]>
