What makes one AI model different from another?

Maker, version, size of the family, settings and access all change what you get.

Facts below come from the datasets on this site, last updated 29 Oct 2026, 14:36 UTC. Interpretation is labelled as ours. We have not tested these products hands-on.

What the evidence shows

• The 77 models here come from 16 makers: Alibaba, Amazon, Anthropic, Baidu, ByteDance, DeepSeek, Google DeepMind, Meta, Mistral AI, Moonshot, NVIDIA, OpenAI, Thinking Machines, Xiaomi, Z.ai (Zhipu AI) and xAI.
• Access as listed by Epoch AI:
  • API access: 35
  • Limited access: 1
  • Not stated: 28
  • Open weights (non-commercial): 1
  • Open weights (restricted use): 1
  • Open weights (unrestricted): 11
• One maker's family can span the ratings. For example, some of Anthropic's listed models, for writing: Claude Fable 5 (Top tier for writing), Claude Fable 5.1 (Top tier for writing), Claude Haiku 5.5 (Behind the leaders for writing), Claude Opus 4.5 (20251101) (Strong for writing), Claude Opus 4.6 (Top tier for writing), Claude Opus 4.7 (Top tier for writing).
• Sources test models at different settings. Each result on a model page shows its setting.

Our interpretation

A name hides several variables: the version, the size of the model within a family, how much “thinking” it is set to use, and how you can reach it. The same name at two settings can behave differently, which is why we record the setting beside every score.

What to do next

Open any model page to see its settings and sources.