Atlas

Models

← All models

Opus 4.6

Anthropic · 2026-02-05

Claude Opus 4.6 is Anthropic's February 2026 hybrid-reasoning large language model for advanced knowledge work, coding, and agentic tasks. It introduced adaptive thinking, allowing the model to decide when deeper reasoning is useful, while retaining optional manual extended thinking. It supports a 1M-token context window.

Benchmark scores

BenchmarkScore
A-Fantasia30.0
A-Fantasia (Valid Attempts with Retries)29.7
A-Fantasia Backwards Spelling17.0
A-Fantasia Chess20.0
A-Fantasia Chess (Valid Attempts with Retries)20.0
A-Fantasia Cube (Valid Attempts with Retries)53.0
A-Fantasia Cube Rotation53.0
A-Fantasia Spelling (Valid Attempts with Retries)16.2
AA-LCR70.7
AA-LCR v1.167.0
AA-Omniscience Index13.7
Agent Red Teaming k=100 (CAIS)49.5
Agent4Sci mean relative improvement20.1
AIME 2024 (avg@8)97.5
AIME 2024-202595.6
AIME 202599.8
AIME 2025 (avg@8)93.8
AIME 202696.7
ALE-Bench1607
AlgoTune1.5
Antidote: Everyday Edition1042
Apex (MathArena)34.5
APEX (Mercor)14.9
Apex Shortlist86.7
APEX-Accounting43.3
APEX-Accounting Pass@14.4
APEX-Agents32.4
APEX-Agents 1.161.8
APEX-Agents 1.1 Pass@146.3
APEX-Agents-AA (avg@3)33.0
Loading Atlas data…