Eseandre Mordi
⬤ Korea Telecom just dropped Mi:dm K 2.5 Pro, a reasoning-focused AI model built specifically for enterprise and telecom applications. The model hit an overall score of 48 on the Artificial Analysis Intelligence Index v3.0, which tests AI systems across 10 different benchmarks covering everything from coding and math to real-world tasks and agentic workflows.
⬤ Where Mi:dm K 2.5 Pro really shines is in agentic tool-use evaluation. On the τ²-Bench Telecom workflow test, it scored 87% — putting it right up there with top-tier models like Gemini 3 Pro Preview and Claude Opus 4.5. This makes it one of the strongest performers specifically for telecom-related agentic tasks.
⬤ The model crushes it in Korean-language capability, scoring 83% on the Global MMLU Lite Multilingual Index for Korean. That's comparable to gpt-oss-120B and Naver's HyperCLOVA X SEED Think, reflecting how Korean tech companies are increasingly optimizing models for their home market. One trade-off: Mi:dm K 2.5 Pro uses a hefty amount of reasoning tokens — around 90 million across the full evaluation suite, similar to EXAONE 4.0 32B.
⬤ Mi:dm K 2.5 Pro isn't publicly available. Korea Telecom plans to integrate it into its own product lineup and deploy it for enterprise and telecom clients rather than releasing it as a general-purpose platform. More details are expected as the rollout continues.
Eseandre Mordi
