Nous Research releases Hermes 4

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Hermes 4 released — corpus expanded to about 5 million samples (60B tokens), centered on validated reasoning traces; <think> hybrid reasoning; 14B/70B/405B weights fully released

Why it matters

An approach that benchmarks lower refusal rates as an alignment metric in their own right. The open community formalizes an alternative axis to safety alignment

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2508.18255

Added in Week 35, 2026 · See that week

On the map — ⑨ Open-Weight Lineage line

  1. Previous stopDeepSeek-V3.1
  2. This recordHermes 4
  3. Next stopKimi K2 Thinking
See it on the map

Same week

xAI releases grok-code-fast-1OfficialOn the Theoretical Limitations of Embedding-Based Retrieval publishedPaperAnthropic and OpenAI publish a mutual alignment evaluation pilotOfficialGemini 2.5 Flash Image (nano-banana)Official

Nearby on the same axis

xAI releases Grok 2 (2.5) weights on Hugging FacePressDeepSeek releases DeepSeek-V3.1 with hybrid thinkingOfficialgpt-oss-120b / 20bOfficialSK Telecom open-sources A.X 3.1Official