Atlas

Models

← All models

HyperNova 60B 2605

Multiverse Computing · 2026-05-06 · 58.7B parameters

HyperNova 60B 2605 is Multiverse Computing's open-weight MoE reasoning model released in May 2026 as an updated version of HyperNova 60B 2602, with stronger coding and general benchmark performance. It was developed from gpt-oss-120b using Multiverse Computing's CompactifAI compression technology, is reported to activate 4.8B parameters per token, and supports configurable reasoning effort, native tool calling, structured outputs, and agentic workflows.

Benchmark scores

BenchmarkScore
AA-LCR31.7
AA-Omniscience Index-56.4
Artificial Analysis Agentic Index6.7
Artificial Analysis Coding Index23.2
Artificial Analysis Intelligence Index17.8
Artificial Analysis Omniscience Accuracy14.2
Artificial Analysis Omniscience Hallucination Rate82.2
Artificial Analysis Openness Index38.9
Capability138.5
CritPt0.9
GDPval-AA v2658
GDPval-AA v2 (normalized)7.9
GPQA Diamond73.3
Humanity's Last Exam (Text-Only)15.1
IFBench (Artificial Analysis)66.5
SciCode (Artificial Analysis)33.0
Terminal-Bench 2.118.4
Terminal-Bench Hard23.5
τ²-bench (Telecom)63.2
𝜏³-Banking5.4
Loading Atlas data…