Every frontier model, side by side.
The newest release from every major AI lab — specs, pricing, abstract-reasoning, science, coding and agentic scores, a value ranking and a cybersecurity posture index. Click any column header to sort.
Compare models →CybersecurityWho’s building what
Ten labs define the frontier and its open-weight tail.
Every latest model
One row per current model — older generations are dropped. AA Index = Artificial Analysis Intelligence Index v4.3.2. Prices per million tokens. Value = Index ÷ blended price (3 parts input : 1 part output). Every cell is filled; EST marks our estimate where a lab hasn’t published a comparable figure.
Compare up to three models
Pick models to see every spec side by side. Best value in each row is highlighted green.
Reasoning, science & agentic coding
No certified AGI test exists; these are the closest public proxies.
Artificial Analysis Intelligence Index
Composite of 10 evals: reasoning, knowledge, coding, math, agentic work.
Terminal-Bench 4.0 — agentic terminal tasks (%)
Multi-step software, sysadmin and data tasks, mostly vendor-reported.
ARC-AGI-2 — novel abstract reasoning (%)
Puzzles that resist memorization; average human ≈ 66%.
GPQA Diamond — graduate-level science (%)
PhD-level, Google-proof multiple-choice questions.
SWE-bench — real GitHub issue fixing (%)
Resolving real software issues end to end.
Cybersecurity posture index
| Model | Secure code | Injection resist. | Attack refusal | Gating | Notes |
|---|
How this page is built
Public sources
Lab release notes, system cards and public leaderboards.
Monthly refresh
New releases replace older versions of the same model line.
No single “best”
Rankings shift by task and reasoning effort.
Estimates marked
Gaps are filled with clearly badged estimates, never silently.
