Compare models

Three things to know first: plan, model, harness

Switching models in Nuo actually involves three things: which Plan you use it through, which Model does the thinking, and which Harness puts it to work. The leaderboards below compare models.

1. Plan

   Where you use it and pay

   What you buy is quota for using models, not the models themselves. A plan usually includes several models, and they aren’t necessarily made by the company selling the plan. For example, besides Alibaba’s own Qwen, Alibaba Cloud Bailian also offers Zhipu’s GLM and DeepSeek’s models. The other way around, the same model may be in several plans (GLM-5.3 is available in Alibaba Cloud Bailian, Zhipu and Volcengine Ark).

2. Pick one
3. Model

   Who does the thinking

   The AI that actually reads the task, works out how to do it and writes the answer is trained by a particular company: Claude by Anthropic, Qwen by Alibaba, GLM by Zhipu, DeepSeek by DeepSeek. A model is about as capable whichever plan you use it through; speed, how much it can read at once, price and quota may differ.

4. Put to work
5. Harness

   Puts the model to work

   The program that connects a model to tools (reading and writing files, running commands, looking things up online, editing docs, operating your computer) and has it finish a job step by step. The same model can do quite differently in a different harness. Nuo itself is a harness: whichever plan and model you choose, Nuo gives it the tools and guides it step by step until the job is done.

So leaderboard scores are measured in each leaderboard’s own test setup (the coding and agent ones also use their own harnesses). In Nuo, a model performs at roughly that level, but not to the exact point.