K2 Think V2: benchmark results
MBZUAI's open sovereign 70B reasoning model, RL-trained on the K2-V2 base. Provider: LLM360. Released 2026-01-27. Access: Open.
Aggregate Rating 1546 ± 25, rank #620 of 1635 rated models, from 67 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA IFBench | 62.79 | Accuracy (%) | 73.7 |
| AA Omniscience - Software Engineering (SWE) - JavaScript | 33.64 | Accuracy (%) | 69.1 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 20 | Accuracy (%) | 63 |
| AA-Omniscience Hallucination Rate | 71.17 | Hallucination Rate (%) | 60.7 |
| AA Omniscience - Software Engineering (SWE) - HTML | 32 | Accuracy (%) | 60.4 |
| SurgiQ - Laparoscopic Surgery | 58.01 | Laparoscopic surgery questions (493): accuracy (%) on SurgiQ | 60.3 |
| AA Omniscience - Software Engineering (SWE) - Rust | 50 | Accuracy (%) | 59 |
| SurgiQ - Critical Care and Emergency | 59.66 | Critical care and emergency surgery questions (590): accurac | 58.8 |
| AA Omniscience - Software Engineering (SWE) - Julia | 12 | Accuracy (%) | 58 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 21.11 | Accuracy (%) | 57.5 |
| SurgiQ - Case-Based | 58.2 | Case-based questions (6,156): accuracy (%) on SurgiQ's four- | 55.9 |
| AA-Omniscience Index - Software Engineering (SWE) - JavaScript | -18.18 | Omniscience Index | 54.1 |
Interactive version: theaggregate.ai/model?slug=k2-think-v2 · How It Works · Data refreshed daily, snapshot 2026-10-15.