Know which AI model is best right now, and what each one is best at.
Every major AI model ranked on more than 40 benchmarks, from real paid work and coding agents to the hardest exams, with what each benchmark measures in plain words. See what jumped with each new release, which model leads each kind of work, and what each maker says its model is good at. Checked every four hours, so a new model is on the board the day it ships.
Free on Explorer. No card.
The model leaderboard today
Real output from Spezia, read from the same data the tool uses. Every line links to its source.
- 1
Claude Fable 5.22,264 · first on 10
- 2
Gemini 4 Carbon2,253 · first on 4
- 3
- 4
- 5
GPT-6 Astra1,927 · first on 3
- 6
- 7
- 8
- Real work
- Agents and tools
- Coding
- Reasoning and knowledge
- Math
- Vision
Kimi K3.1 Pro
What jumped with the newest releases
ARC-AGI-3▲ 2.6×30.2% → 79.3% vs Claude Opus 5Remote Labor Index▲ 2.3×17.9% → 41.3% vs Claude Fable 5.1SWE-Bench Pro▲ +13 pts80% → 93% vs Claude Fable 5
Doubao Seed Evolving2026-11-02
GPQA Diamond▲ +3.5 pts89.6% → 93.1% vs Doubao Seed 2.1 Turbo
Every benchmark ranked and charted, and each model's profile, are inside.
Continue free →Free account, no card.
123 models rated on 44 benchmarks in use. Independent results from Epoch AI and LMArena (CC BY 4.0) and makers' published numbers collected by LLM Stats, never ranked against each other.
What you get
The best model right now
One overall ranking from every benchmark two models share, so you see at a glance which model leads, by how much, and which leads each kind of work.
What jumped with each release
Each new model against the one it replaced, benchmark by benchmark, shaded by how big the gain is, so a doubling stands out.
Every benchmark explained
What each of more than 40 benchmarks asks a model to do, why it matters and how it is scored, with the leader and the record over time.
How it works
- 1Open the leaderboard and see the best model right now.
- 2Pick a benchmark to rank every model on it and see the record over time.
- 3Open a model for where it leads, where it trails and what its maker says about it.
Who uses it
Questions about Model leaderboard
How is the overall ranking worked out?
Like a chess rating. Every benchmark two models both have a result on counts as a game won by the higher score, and beating a strong model counts for more than beating a weak one. Tests of real work, agents and the hardest reasoning count fully; tests the best models have nearly maxed out count less.
Where do the numbers come from?
Independent results from Epoch AI and LMArena, and the numbers makers publish in their own announcements and system cards, as collected by LLM Stats, each linked to its source. A maker's own number is marked as such and never ranked against an independent lab's.
How fast is a new model added?
The board is checked every four hours. A new model appears the day it is announced or offered, and its results fill in as they are published.
What is free?
The whole board: the ranking, every benchmark and chart, what jumped with each release, and the profiles of the top 5 models. Plus, $29 a month, opens every model's profile.
Often used with
Tooling guide
NewFreeThe best model and tool for each kind of work, rewritten weekly.
See it liveAI Progress
FreeReleases, benchmarks and an outlook for the next 3 to 24 months.
See it liveSoftware watch
NewFreeNew models, features, prices and limits, the day they ship.
See it live
Know which AI model is best right now, and what each one is best at.
Free on Explorer, with 5 questions a month to Ask Spezia. No card.