AI Model Tracker — LLM benchmarks
& leaderboard.

GPT-6 Astra currently leads 2 of 10 capability axes. The five leading models on each, on one honest scale — composite scores are weighted averages over each axis's audited benchmark basket.

10 AXES·480 MODELS·44 BENCHMARKSUPDATED OCT 9 · 06:19 UTCScore basis

Tier 1 — Established axesHigh coverage, audited harnesses, narrow confidence.

Reasoning

480 MODELS SCOREDDEAD HEAT

Composite /100 · Most authoritative basis

1OGPT-6 Astra93.1

2AClaude Fable 5.192.2−0.9

3AClaude Opus 590.9−2.2

4OGPT-5.6 Sol90.7−2.4

5AClaude Fable 589.9−3.1

LeadingFieldWINDOW 89.9–93.1 / 100FULL STANDINGS →

Coding

259 MODELS SCOREDCONTESTED

Composite /100 · Most authoritative basis

1AClaude Fable 583.0

2AClaude Fable 5.179.3−3.7

3AClaude Opus 578.2−4.9

4OGPT-5.6 Sol77.9−5.1

5AClaude Opus 5.575.4−7.6

LeadingFieldWINDOW 75.4–83.0 / 100FULL STANDINGS →

Agentic

325 MODELS SCOREDCONTESTED

Composite /100 · Most authoritative basis

1AClaude Opus 585.2

2AClaude Fable 582.1−3.2

3OGPT-5.6 Sol81.6−3.6

4KKimi K380.5−4.8

5AClaude Fable 5.179.0−6.2

LeadingFieldWINDOW 79.0–85.2 / 100FULL STANDINGS →

Safety

DATA-THIN

Composite /100 · Most authoritative basis

Scale's SEAL safety boards (MASK honesty, Fortress adversarial risk, PropensityBench misuse propensity) now flow daily — 46 models measured on two or more. The cross-vendor safety ranking is parked pending calibration research; board scores appear on each model's page, labeled, with direction stated (Fortress and PropensityBench count risk — lower is better).

Tier 2 — Emerging axesSufficient signal to rank; bands wider where coverage thins.

Long Context

372 MODELS SCOREDTHIN BASKETCONTESTED

Composite /100 · Most authoritative basis

1OGPT-5.6 Terra76.4

2GGemini 3.7 Flash75.4−1.0

3QQwen3.8 Max Preview75.0−1.4

4KKimi K373.5−3.0

5SStep 5 Preview73.4−3.1

LeadingFieldWINDOW 73.4–76.4 / 100FULL STANDINGS →

Math

80 MODELS SCOREDCONTESTED

Composite /100 · Most authoritative basis

1OGPT-6.1 Sol80.5

2OGPT-6 Astra79.3−1.2

3AClaude Opus 5.577.8−2.7

4OGPT-6 Sol76.4−4.1

5AClaude Fable 5.175.6−4.9

LeadingFieldWINDOW 75.6–80.5 / 100FULL STANDINGS →

Multimodal

190 MODELS SCOREDCONTESTED

Composite /100 · Most authoritative basis

1GGemini 3 Pro75.0

2BSeed 2.1 Pro Preview72.9−2.1

3KKimi K2.570.1−4.9

4GGemini 3.1 Pro68.5−6.5

5QQwen3.5 122B A10B68.0−7.0

LeadingFieldWINDOW 68.0–75.0 / 100FULL STANDINGS →

Multilingual

DATA-THIN

Composite /100 · Most authoritative basis

Scale's MultiNRC multilingual board now flows (36 models, 8 vendors — French, Spanish and Chinese). A single-board three-language ranking is parked pending research; multilingual scores appear on model pages. Re-opens with a broader source or the research pass.