Apertus 1.5 extends open models to multimodal reasoning with 262K context
swiss-ai · hf · 2026-10-15
Swiss public-institution team releases Apertus 1.5, extending the fully open Apertus family from text to multimodal reasoning. Continued pretraining adds 4T tokens (8B) / 2T tokens (70B), image and audio understanding, and a 4x context extension to 262,144 tokens, with architecture unchanged — multimodality enters through the tokenizer. Post-training includes SFT, long-context finetuning, RLVR, and preference alignment; thinking is a chat-template flag that lifts math reasoning by up to 20 points on the 8B. The 8B leads fully open peers on knowledge, instruction following (beating all compared models including open-weight ones), and culture benchmarks, and is competitive on image/audio understanding, strongest on document understanding and low-resource speech recognition. Training uses only openly available data with retroactive crawler opt-out; weights, recipes, and eval suites ship under permissive licenses.
More from Models
- Fish Audio Drama 3 lands in vLLM-Omni: 128 concurrent streams on one H200 — vllm_project · 2026-10-15
- Why is GPT-6 Astra ranked so low on LM Arena? Outscored by GPT-5.6 and 6.1 — AreaExact7824 · 2026-10-15
- Stance test heat map shows Sonnet 5.5 resists framing manipulation other models succumb to — teortaxesTex · 2026-10-15
- Argon (Medium) burned 85% of a 5-hour usage limit in just 5 minutes — Avalunne · 2026-10-15
- Score vs cost chart puts Gemini 4 Argon on the Pareto frontier, Kimi K3 far off it — Hesamation · 2026-10-15
- OpenRouter data shows Anthropic's Opus 5.5 clawing back share from OpenAI within days — menhguin · 2026-10-15