Sulat.com
AI models
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
Infomaniak logo

Model details

Apertus v1.5 70B

Apertus v1.5 70B is the latest iteration of the Apertus family from Swiss AI, distributed as an open-weight model under the Apache-2.0 license. Community listings tag the architecture as multimodal and image-text-to-text, reflecting the base model's ability to handle visual and textual inputs alongside standard language tasks. The base checkpoint on Hugging Face is published under the swiss-ai/Apertus-v1.5-70B repository, and quantized derivatives such as onprem-ai/Apertus-v1.5-70B-NVFP4 preserve the lineage while offering more efficient deployment footprints.

For organizations prioritizing efficient single-node inference, the NVFP4-quantized variant shrinks the original 135 GiB BF16 weights to roughly 48 GiB while claiming substantial decode speedups and near-FP8 quality retention. This format is recommended for production on a single 96 GiB GPU, leaving headroom for large KV caches and multimodal encoder memory. An FP8 sibling checkpoint at 71 GiB is also available for teams that prefer a less aggressive quantization trade-off, giving integrators flexibility between raw fidelity and resource efficiency.

Infomaniakswiss-ai/Apertus-v1.5-70Bbeta
Open provider pageProvider docs

Quick Info

Powered by
Provider
Infomaniak
Model key
swiss-ai/Apertus-v1.5-70B
Release date
Jul 24, 2026
Last updated
Aug 1, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost
$0.87
Output token cost
$3.10

Limits

Input tokens
100,000 tokens
Output tokens
8,192 tokens
Context window
100,000 tokens
News search byUse it for free

Latest news about Apertus v1.5 70B

Infomaniak

CoverageRelease Notes

Nathan Lambert and Florian Brand's Interconnects #23 essay (Aug 2, 2026), summarized via ai-tldr.dev, catalogs roughly ten recent open-model releases and explicitly names Swiss AI's Apertus v1.5 70B alongside Thinking Machines' Inkling, Tencent's Hy3, Poolside's Laguna S 2.1, Meituan's LongCat-2.0, AMD's Instella-MoE, While the excerpt offers no technical specs (no benchmarks, context length, license, or release notes for Apertus v1.5 70B), it provides useful independent confirmation that the model shipped as a recent open-weights release from Swiss AI and is being tracked by a respected open-AI commentator. The primary technical an

Infomaniak

Coverage

The Swiss AI Initiative's official Hugging Face organization page lists swiss-ai/Apertus-v1.5-70B as a 72B-parameter Image-Text-to-Text model with multimodal input and reasoning support, last updated about seven days before the current date. The model card shows 17.8k downloads and 104 likes, and the associated Apertus Recent activity on the page confirms ongoing community and developer engagement with the 70B variant: a LEXam-hard evaluation result was added to swiss-ai/Apertus-v1.5-70B about seven days ago, a "Future architecture question" thread is active on the same model, and a separate user contributed experimental llama.cpp su

Videos about Apertus v1.5 70B