by Sarvam AI
Open-weights decision model on Sarvam-105B that returns probabilities for fixed answer options in one forward pass.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Sarvam-Prism is a decision model from Sarvam AI, built on Sarvam-105B. The user supplies a context (state) and one or more questions, each with a fixed set of possible answers. The model returns a probability for every answer in a single forward pass. It does not generate text, so the result is always one of the supplied options.
Three question types are supported: choice (probabilities per option key), yes_no (probability of yes) and score (ordered levels, with an expected level index and per-level probabilities). All questions in a request are scored together, in batches, and each is scored independently. Use cases include ticket routing, urgency classification and sentiment scoring. The model card lists support for English and Indian languages including Hindi, Bengali, Tamil and Telugu. Sarvam reports a median latency of 44 ms per request on one NVIDIA B200.
The weights are released under Apache-2.0 and are about 106 GB. They are stored in a custom FP8 layout and must be loaded with the provided prism.py. They cannot be loaded with AutoModelForCausalLM or served by general-purpose inference servers. Self-hosting needs one FP8-capable GPU with about 120 GB of free memory, such as a B200 or H200. A hosted API is also available from Sarvam.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-weights decision model built on Sarvam 105B. Given context and questions with fixed answer options, it returns a probability for each answer in one forward pass without generating text. It needs about 106 GB of disk and a B200/H200-class GPU.
Robot control system that turns sensor readings, images and instructions into movement actions, using combined signals such as force, velocity, contact and friction. It shows a higher success rate than comparable algorithms, and code is released.
Closest neighbours by meaning, not just by category.
Model
Open 7.39B dense model for math reasoning and agentic search, with a 256K context and the full training pipeline released.
Model · Microsoft
Microsoft's small decision-scoring model returning calibrated probabilities for routing, classification and rubric grading.
Model · Meituan
Meituan's open-weights 1.6T-parameter mixture-of-experts LLM (48B active), aimed at agentic coding and reasoning tasks.
Model · Meta
Meta's open-weight 816B-parameter mixture-of-experts model (42B active, 1M context) in the Muse model family.
Model
Open-weight non-autoregressive System 1 decision model returning calibrated probabilities in one forward pass across 100+ languages.
Model · Rednote
Open MoE language model from Rednote (Xiaohongshu) with 208B total and 16B active parameters.