Sarvam AI releases open-weight Sarvam-105B-1711, a 105B MoE with 10.3B active parameters
Sarvam AI released Sarvam-105B-1711 on November 22, 2026, an open-weight checkpoint aimed at customer support, voice agents and other high-volume enterprise work. It ships as BF16 weights for further training and a 4-bit NVFP4 version designed to run on one workstation GPU.
Sarvam AI released Sarvam-105B-1711 on November 22, 2026, a new open-weight checkpoint aimed at customer support, voice agents and other high-volume enterprise work, according to a feed summary from Unite.AI.
Release details
- Model: Sarvam-105B-1711
- Developer: Sarvam AI
- Release date: November 22, 2026
- Architecture: The model card identifies the same 105-billion-parameter mixture-of-experts base used by Sarvam-105B, with 10.3 billion active parameters.
- Post-training: The new checkpoint adds post-training for tool calling.
Formats
The release includes two sets of weights:
- BF16 weights, provided for further training.
- A compact 4-bit NVFP4 version, designed to run on one workstation GPU.
Target use
Sarvam AI positions the checkpoint for customer support, voice agents and other high-volume enterprise work.
What this means
Sarvam-105B-1711 keeps the base architecture of Sarvam-105B and changes the post-training, with tool calling added. That fits the stated focus on enterprise agents, where models call external tools as part of a workflow.
The two formats serve different users. The BF16 weights give teams a starting point for their own further training, while the NVFP4 version is designed for deployment on a single workstation GPU. With 10.3 billion of the 105 billion parameters active, the mixture-of-experts design is built so that only part of the model runs for each token.
Related Articles
Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context
Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.
Baseten releases open-weight bison-1 and groundhog-1 decision models; bison-1 scores 76 on AA Decision Index v2.1
Baseten released bison-1 and groundhog-1, two open-weight decision models built on Gemma 4. Bison-1 scores 76 on the AA Decision Index v2.1, and both models support 128k context and image input.
Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license
Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.
KAIST MLP Lab releases KORMo-Web-11B: open LLM, VLM and web agent scoring 30.1 on Korean Online-Mind2Web
KAIST MLP Lab has released KORMo-Web-11B, an 11B-parameter bilingual model that combines text, vision and browser-agent capabilities in one set of weights under Apache-2.0. On 612 Online-Mind2Web tasks it scores 29.0 in English and 30.1 in Korean, according to the lab.
Comments
Loading...