News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
Solar Open 100B

Solar Open 100B

102B parameter Mixture-of-Experts model by Upstage

Solar Open 100B is a large language model developed by Upstage with 102 billion total parameters. It uses a Mixture-of-Experts design to activate only 12 billion parameters per token, balancing speed and knowledge. Trained on 19.7 trillion tokens, it supports reasoning and agent tasks.

Share on WhatsApp

  • 102B total parameters with 12B active per token
  • Trained on 19.7 trillion tokens from scratch
  • Uses 129 experts in Mixture-of-Experts design
  • Scores 88.2 on English MMLU benchmark
  • Licensed under Upstage Solar License for weights

What it is

Solar Open 100B is a flagship large language model from Upstage. It contains 102 billion parameters in total but uses a Mixture-of-Experts architecture to activate only 12 billion during inference. This approach aims to provide enterprise-grade performance in reasoning and instruction following while maintaining cost efficiency. The model was trained from scratch on a massive dataset of 19.7 trillion tokens. It is designed for transparency and customization within the open-source community.

How it came about

Upstage developed the model entirely from scratch, utilizing NVIDIA B200 GPUs for training. The technical report was published in January 2026. The release includes both the model weights and supporting code, with the weights governed by a specific license. An official INT4 quantized version was also made available through NotaAI to facilitate broader access and deployment.

Inside the model

The architecture relies on a Mixture-of-Experts system with 129 experts in total. For each token processed, the model routes data to the top eight experts from a set of 128 routed options, plus one shared expert. This structure allows the model to leverage the depth of a large parameter count while keeping computational demands lower. Training involved 19.7 trillion tokens, ensuring broad coverage across various domains. The code is licensed under Apache 2.0, while the weights use the Upstage Solar License.

What it can do

The model demonstrates strong performance in Korean benchmarks, scoring 73.0 on KMMLU and 84.4 on KorMedMCQA. In English tests, it achieved 88.2 on MMLU and 91.7 on AIME 2024 for math. Code generation skills reached 74.2 on LiveCodeBench. It also shows capability in agentic tasks, with scores such as 59.3 on Tau² Retail. These results position it competitively against other large models in both general knowledge and specialized reasoning tasks.

How to use it

Users can access the model weights via Hugging Face under the Upstage Solar License. The accompanying code is available under the Apache License 2.0. Deployment is recommended using Docker with vLLM, which supports specific parsers for tool calls and reasoning. Alternatively, users can install from source for custom modifications. Minimum hardware requirements include four NVIDIA A100 GPUs with 80GB memory.

Questions people ask about Solar Open 100B

How many parameters does Solar Open 100B have?

The model has 102 billion total parameters. However, due to its Mixture-of-Experts architecture, only 12 billion parameters are active for each token processed during inference.

What license covers the Solar Open 100B weights?

The model weights are licensed under the Upstage Solar License. The associated code files, such as Python scripts and JSON configurations, are released under the Apache License 2.0.

What hardware is needed to run this model?

The minimum requirement is four NVIDIA A100 GPUs with 80GB of memory. For optimal performance, especially with Docker deployment, eight GPUs are often recommended.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Solar Open 100B yet, so the facts on this page come from the sources listed above: huggingface.co/upstage/Solar-Open-100B. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.