What is Arena Elo — crowd rankings in plain language

Arena Elo is a rating from pairwise human (or human-like) votes: model A vs model B on the same prompt, winner takes points. It is the same family of math as chess ratings. It is not a certification that a model is accurate, safe, or good at your job.

AI Benchmark Hub maps LMArena text-leaderboard Elo into a 0–100 quality index so you can blend it with cost and context. When Arena is down, rows still appear with a blank quality cell. That is better than inventing a number.

Highest quality index right now

Live OpenRouter pricing and Arena quality, cached about an hour. Not a static blog table.

ModelProviderBlend $ / 1MQualityContext
Qwen3.8 Max (0902)Qwen4.0095.11,000,000
Claude Opus 5Anthropic15.0093.41,000,000
Claude Fable 5Anthropic30.0091.11,000,000
GLM 5.3 FlashZ.ai0.1689.71,048,576
Qwen3.8 27BQwen1.7189.01,000,000

Pairwise votes, not absolute truth

Elo moves when the crowd prefers A’s answer to B’s. The crowd is not your customers. English chat, a bit of coding, a bit of roleplay — whatever people paste into an arena. A model can rank high and still fail your schema or your language.

Why we normalize to 0–100

Raw Elo numbers are awkward to mix with dollars and token counts. We min-max scale the scores we have into a quality index. The index is comparable across models on a given fetch, not an absolute IQ.

Matching names is imperfect

OpenRouter ids and Arena names do not always align. We use provider/model matching heuristics. Some rows will lack a quality index even though they are famous. That is a matching miss, not a claim that quality is zero.

What to use instead for production

Your eval set. Use Elo to avoid obviously weak models and to pick a shortlist. Then measure task success. The how-to-choose guide is the process; Elo is one input.

Other arenas

There are coding arenas, vision arenas, and private ones. We currently ingest the text board. Do not read our quality index as a SWE-bench score.

FAQ

Is a higher Elo always better?

Better at winning crowd votes on that board. Not always better at your task or cheaper.

Why is my favorite model missing a quality index?

We couldn’t match it to an Arena row, or Arena was unavailable. The model can still be ranked by cost and context.

Do votes on this site change Elo?

They help you decide. Global quality index comes from the public LMArena text leaderboard we fetch.

Where is this documented?

On /about (methodology) and in the quality card on each model page.
what is Arena EloLMArena Elo explainedChatbot Arena rankingLLM quality index
Read full methodology

Related guides

Also try: LLM leaderboard, compare tool, live arena.