AI Model Rankings 2026
Every category of AI model ranked in one place: large language models, image generators, video generators, audio and voice models, and coding tools. Rankings use arena ELO scores, the ratings built from blind head-to-head comparisons voted on by real users. Models that never entered an arena carry no score and are listed separately rather than given an invented one.
LLM
Large Language Models
15 models ranked
#1 Gemini 4 Argon →
Image
Image Generation Models
10 models ranked
#1 GPT-Image 2.5 Sunburst →
Video
Video Generation Models
12 models ranked
#1 Gemini Omni-1.1 Flash →
Code
AI Coding Tools
5 models ranked
#1 GPT-6 Astra →
Top 15 models overall (arena ELO)
Across every category, the highest-rated models by arena ELO score.
| # | Model | Company | Category | ELO |
|---|---|---|---|---|
| 1 | GPT-6 Astra | openai | Code | 1,789 |
| 2 | GPT-6.1 Sol | openai | Code | 1,759 |
| 3 | Claude Sonnet 5.5 | anthropic | Code | 1,709 |
| 4 | GPT-6 Sol | openai | Code | 1,689 |
| 5 | Kimi K3 | moonshot | Code | 1,658 |
| 6 | Gemini 4 Argon | LLM | 1,533 | |
| 7 | Gemini Omni-1.1 Flash | Video | 1,516 | |
| 8 | Gemini Omni-Flash | Video | 1,513 | |
| 9 | Claude Opus 5.5 | anthropic | LLM | 1,512 |
| 10 | Claude Fable 5.1 | anthropic | LLM | 1,511 |
| 11 | Claude Opus 5 | anthropic | LLM | 1,506 |
| 12 | Claude Opus 4.6 | anthropic | LLM | 1,504 |
| 13 | Gemini 3.8 Flash | LLM | 1,496 | |
| 14 | Flux 3 Video | bfl | Video | 1,493 |
| 15 | Claude Fable 5 | anthropic | LLM | 1,492 |
How these rankings are built
ELO is a relative rating. It is produced by asking users to compare two models head to head without being told which is which, then converting the win rate into a single number. A higher score means the model wins more often against the models it faced — it does not measure speed, price, context window, or availability in your country.
Arena rankings only include models that compete in the arena. Several widely used commercial products never do, so they appear in these tables under “no arena score yet” and are ordered by adoption instead. That gap is deliberate: a rating invented for a model that never competed would be worth nothing to you.
The visual side of the same data — bubbles sized by adoption and coloured by category — is on the AI Bubbles homepage, and you can embed it on your own site for free.