---
title: "AI model leaderboard: blind-vote rankings across 17 boards"
description: "Rankings come from blind votes by real people: the same prompt goes to two anonymous models, and the user picks the better answer. The top model often changes from board to board."
canonical: https://pirouter.ai/rankings
last-updated: 2026-10-15
---

# AI model leaderboard: blind-vote rankings across 17 boards

> Rankings come from blind votes by real people: the same prompt goes to two anonymous models, and the user picks the better answer. The top model often changes from board to board.

## Text

- [Overall leaderboard](https://pirouter.ai/rankings/overall): Overall rankings of text models across math, coding, creative writing and other open-ended tasks. Gemini 4 Argon (1525), Claude Opus 4.6 (1506), Claude Fable 5 (1505) · 2026-09-30
- [Coding leaderboard](https://pirouter.ai/rankings/coding): Rankings on programming prompts: writing, debugging, explaining and refactoring code. Gemini 4 Argon (1559), Claude Fable 5 (1552), Claude Opus 4.6 (1551) · 2026-09-30
- [Math leaderboard](https://pirouter.ai/rankings/math): Rankings on math derivation and calculation prompts. Gemini 4 Argon (1528), Claude Opus 5 (1523), Claude Fable 5 (1522) · 2026-09-30
- [Creative writing leaderboard](https://pirouter.ai/rankings/creative): Rankings on stories, copy and poetry. Gemini 4 Argon (1522), Claude Opus 5.5 (1515), Claude Fable 5 (1503) · 2026-09-30
- [Instruction following leaderboard](https://pirouter.ai/rankings/instruct): Rankings on answering within format, length and step constraints. Gemini 4 Argon (1529), Claude Opus 5.5 (1514), Claude Opus 4.6 (1514) · 2026-09-30
- [Multi-turn leaderboard](https://pirouter.ai/rankings/multiturn): Rankings on multi-turn conversations. Gemini 4 Argon (1552), Muse Spark 1.2 (1522), Claude Opus 4.6 (1519) · 2026-09-30
- [Long prompts leaderboard](https://pirouter.ai/rankings/longctx): Rankings on long-input prompts. Gemini 4 Argon (1544), Claude Opus 5.5 (1528), Claude Opus 4.6 (1524) · 2026-09-30
- [Chinese leaderboard](https://pirouter.ai/rankings/chinese): Rankings on prompts written in Chinese. Gemini 4 Argon (1588), Claude Fable 5.1 (1573), Claude Opus 5.5 (1572) · 2026-09-30

## Coding & agents

- [Web dev leaderboard](https://pirouter.ai/rankings/webdev): Rankings on building a working web page from a one-sentence request. Claude Opus 5.5 (1815), GPT-6 Astra (1788), Claude Sonnet 5.5 (1786) · 2026-10-01
- [Agents leaderboard](https://pirouter.ai/rankings/agent): Rankings on finishing multi-step tasks in real agent sessions; scores are net improvement over a baseline (percentage points). Claude Fable 5.1 (+14.6), Claude Opus 5.5 (+13.8), GPT-6 Astra (+12.2) · 2026-09-30

## Multimodal

- [Vision leaderboard](https://pirouter.ai/rankings/vision): Rankings of multimodal models that reason over images: charts, screenshots, OCR and photos. Claude Fable 5 (1310), Qwen3.8 (1301), Claude Opus 4.6 (1299) · 2026-09-28
- [Documents leaderboard](https://pirouter.ai/rankings/document): Rankings on document analysis and long-content reasoning. Claude Fable 5 (1499), Claude Opus 4.6 (1495), Claude Opus 4.7 (1494) · 2026-09-13
- [Search leaderboard](https://pirouter.ai/rankings/search): Rankings of models with integrated web search. GPT-5.6 Sol (1256), Claude Fable 5 (1229), GPT-5.5 Search (1224) · 2026-08-24

## Generation

- [Text to image leaderboard](https://pirouter.ai/rankings/t2i): Rankings of text-to-image models. GPT-image 2.5 Sunburst (1424), GPT-image 2.5 Flare (1401), GPT-image 2 (1383) · 2026-09-24
- [Image editing leaderboard](https://pirouter.ai/rankings/imgedit): Rankings of image editing models. GPT-image 2.5 Sunburst (1522), GPT-image 2.5 Flare (1479), GPT-image 2 (1461) · 2026-09-30
- [Text to video leaderboard](https://pirouter.ai/rankings/t2v): Rankings of text-to-video models. Gemini Omni 1.1 Flash (1516), Gemini Omni Flash (1513), Flux 3 Video (1493) · 2026-09-22
- [Image to video leaderboard](https://pirouter.ai/rankings/i2v): Rankings of image-to-video models. Minimax H3 (1495), Gemini Omni 1.1 Flash (1489), Wan3.0 (1480) · 2026-09-22

## Where the data comes from

### Blind votes

Users send the same prompt to two anonymous models and vote after reading both answers. Votes become scores through a Bradley-Terry model; scores compare only within one board. Text and vision boards use style control to reduce the effect of answer length and formatting.

### Rank spread

Every score has a 95% confidence interval. A rank spread of 2–12 means the model ranks somewhere between #2 and #12. Models whose intervals overlap may swap places.

### Updates and reasoning tiers

Updated when new models ship or the rankings change; each chart shows its board’s update date. Only the best-scoring reasoning tier of a model is kept, and it is labelled.

Data: [LMArena leaderboard dataset](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset), used under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/); filtered and reasoning tiers merged, scores unchanged.
View the original board: https://arena.ai/leaderboard