Google’s latest model, Gemini 3, has reached the number one position on the Icelandic LLM Leaderboard, curated by Miðeind.

The model achieved an average score of 88.07%, a significant jump from the previous best of 82.6% held by Gemini 2.5. This secures the top two spots for Google.

Most notably, Gemini 3 set a new standard in WikiQA-IS, a benchmark for Icelandic culture and history, scoring 64.82% compared to the previous high of 52.66%. Performance in grammatical error detection (GED) also reached 81%, surpassing the previous record of 75% held by GPT-5.1.

These results suggest that current benchmarks are becoming saturated and harder tests are needed. However, it is excellent news that Icelandic language capabilities have become so advanced that we now need to curate even more difficult benchmarks.

👉 View the leaderboard here: Icelandic LLM Leaderboard
ℹ️ Read more about the leaderboard: An Icelandic leaderboard for large language models