MBZUAI's open-weights K2 Horizon 375B jumps 30 points to 47 on the Intelligence Index
ArtificialAnlys · x · 2026-09-03
UAE's MBZUAI Institute of Foundation Models released K2 Horizon 375B A23B, an open-weights MoE model (375B total / 23B active) scoring 47 on the Artificial Analysis Intelligence Index — a 30-point jump over its predecessor K2 Think V2 (17) and on par with MiniMax-M3 (45).
- Architecture: 375B/23B-active MoE, context extended from 262K to 512K tokens, text-only; pricing and license details TBD.
- Strong agentic, weaker deep reasoning: leads GDPval-AA (Elo 1430 vs MiniMax-M3's 1380) and τ³-Banking (34.2% vs 15.3%), but trails on GPQA Diamond (87.3% vs 92.9%) and Humanity's Last Exam (32.0% vs 39.0%).
- Very low hallucination: it attempts only 40% of AA-Omniscience questions, declining the rest — 26% hallucination rate (predecessor: 71%) with accuracy essentially unchanged at 18%.
- K2 Think V2 ranked among the most open models; K2 Horizon is expected to join the Openness Index once docs ship.
Related event: MBZUAI's IFM Releases K2 Horizon: Six Fully Open Models from 0.9B to 375B(30 posts)→
More from Models
- Users want open-format export from Claude Docs as charts only save as images — BlackHC · 2026-10-20
- EMNLP study: LLM safety interventions drop up to 4.5x when distress is wrapped in delusion — MaartenSap · 2026-10-20
- Cybersecurity-focused VLM OrcaCyber-Nano-1.0 trends on Hugging Face — orcarouter · 2026-10-20
- 0.8B local model with 2,000 labels beats Jev's frontier API on banking data — kalyan_kpl · 2026-10-20
- New TTS benchmark tests 26 models across 30,564 requests: fastest at 46ms — mattturck · 2026-10-20
- Agents handle unreleased language Jai flawlessly, even compile-time execution — lucasmeijer · 2026-10-20