Swiss team's open Apertus 1.5 adds multimodal input and 262K context, with all training checkpoints public
lmoroney · x · 2026-10-17
The Swiss AI team behind Apertus released Apertus 1.5: 8B and 70B models (Apache 2.0) now accept image and experimental audio input, add an optional thinking mode, and extend the context window 4x to 262,144 tokens. Images and audio enter as discrete tokens in the same sequence as text (a 512×512 image costs 1,024 tokens; audio runs 40 tokens/sec).
Notably, they published checkpoints from every training stage (long-context base, SFT, RLVR, DPO, final), plus data sources, recipes, and rerunnable evals on a public lm-evaluation-harness fork. The author suggests a classroom exercise: run five prompts through each stage and have students observe how answers shift — a hands-on way to teach post-training.
Related event: Apertus 1.5 Open-Source Release Adds Image, Audio and 262K Context(2 posts)→
More from Models
- Will small models' quadrillions of tokens outquality Opus 5.5? One bet says soon — LesaunH · 2026-10-17
- AI4Bharat and NVIDIA release 2.8T-token open dataset and evals for all 22 Indian languages — CatAstro_Piyush · 2026-10-17
- Physics research benchmark: Astra tops with 98% pass rate, GLM trails at 42% — teortaxesTex · 2026-10-17
- Gemini Tasks gated away from Pro users, screenshot shows upgrade prompt — EnvironmentalGur4953 · 2026-10-17
- Defiance Repack Ships Uncensored Qwen3.8-27B With 40 Switchable Persona Modes — Distinct-Pie2389 · 2026-10-17
- NVIDIA releases multimodal reranker on Hugging Face for multilingual vision RAG — solyarisoftware · 2026-10-17