Live index · 52 models · Updated Oct 2026
RSS/API/DOCS

ModelRegistry

v2026.9
Add

Frontier AI Leaderboard

Editorial snapshots comparing foundation models domain by domain, reviewed October 2026. Leaders hold the highest published score in each comparison - not a universal “best model” title. Methodology

Reasoning & STEM Intelligence

Models with adaptive test-time compute, deep chain-of-thought, and autonomous multi-turn reasoning.

Comparison leaderClaude Fable 5.1 / GPT-6 Astra / Grok 4.6
01Claude Fable 5.1
$10/M
02GPT-6 Astra
$10/M
03ChatGPT 5.6
$3/M
04Grok 4.6
$2/M

How this is judged: Compares lab-published reasoning and STEM scores (Terminal-Bench, SWE-bench family, GPQA Diamond, AIME) across indexed models. The named leader holds the highest published figure in this comparison, not a universal 'best model' title. Reviewed 2026-10-07. swebench.comgithub.comartificialanalysis.ai

Agentic Software Engineering & Coding

Frontier performance on long-horizon code refactoring, Terminal-Bench execution, and tool orchestration.

Comparison leaderClaude Fable 5.1 (Terminal-Bench 52.6%, lab-reported) / Gemini 3.8 Flash
01Claude Fable 5.1
$10/M
02ChatGPT 5.6
$3/M
03Gemini 3.8 Flash
$0.75/M
04Grok 4.6
$2/M

How this is judged: Compares lab-published software-engineering scores (SWE-bench family incl. DeepSWE, Terminal-Bench, SWE-Bench Verified/Multilingual) across indexed models. Figures are lab-reported; harnesses and versions differ between vendors. Reviewed 2026-10-07. swebench.comgithub.com

High-Volume Value

Models developers actually route at massive scale on OpenRouter - observed usage share in production, not benchmark scores.

Comparison leaderDeepSeek V4 Flash 0731 (49.9T tokens/30d observed) / MiMo-V2.5 (leading coding share)
01GLM 5.3 Flash
$0.35/M
02MiniMax M3
$0.3/M
03MiMo-V2.5
$0.14/M
04DeepSeek V4 Flash 0731
$0.09/M

How this is judged: Compares public inference volume and price-performance (OpenRouter weekly token rankings, provider list pricing) for high-throughput models. Volume reflects observed usage share, not quality. Reviewed 2026-10-07. openrouter.ai

Open Weights & Self-Hosting

Publicly downloadable weights under open and community licenses for enterprise sovereignty and private clusters.

Comparison leaderQwen3.8 2.4T A95B (2.4T MoE) / Llama 4 Maverick (128E MoE)
01DeepSeek V4.1 Flash
$0.3/M
02DeepSeek V4 Flash Vision Exp
$0.12/M
03Muse Spark 1.3
$0.05/M
04Llama 4 Maverick (128E)
$0.45/M
05Llama 4 Scout (16E)
$0.2/M

How this is judged: Ranked live from current registry figures - no editorial shortlist, no fixed evaluation date.

Context Window Capacity

Maximum tokens accommodated in a single inference session without losing retrieval precision or needle recall.

Comparison leaderLlama 4 Scout (16E) (1.31M) / MiMo-V2.5 (1.10M) / MiMo-V2.5 Pro (1.10M)
01Llama 4 Scout (16E)
$0.2/M
02MiMo-V2.5
$0.14/M
03MiMo-V2.5 Pro
$0.46/M
04GPT-6 Astra
$10/M
05ChatGPT 5.6
$3/M

How this is judged: Ranked live from current registry figures - no editorial shortlist, no fixed evaluation date.

Inference Cost & Value

Lowest input/output pricing per 1M tokens combined with near-frontier intelligence for production applications.

Comparison leaderJev ($0.042/M) / Muse Spark 1.3 ($0.05/M) / DeepSeek V4 Flash 0731 ($0.09/M)
01Jev
$0.042/M
02Muse Spark 1.3
$0.05/M
03DeepSeek V4 Flash 0731
$0.09/M
04DeepSeek V4 Flash 0423
$0.09/M
05GPT-6 Luna
$0.1/M

How this is judged: Ranked live from current registry figures - no editorial shortlist, no fixed evaluation date.

This leaderboard is an editorial snapshot, not an objective ranking. Scores are lab-published figures that vendors measure with different harnesses and versions; usage stats reflect observed public routing share. Full rules: /methodology · Record-level sources: /changelog