Atlas

Models

← All models

Claude Haiku 5.5

Anthropic · 2026-10-07

Claude Haiku 5.5 is Anthropic's October 7, 2026 model for high-volume, latency-sensitive work. It supports text and image inputs, adaptive thinking with adjustable effort, a one-million-token context window and up to 128K output tokens. Adaptive thinking defaults to medium effort. The official migration guide says it has no server-side model fallback. Its June 2026 knowledge cutoff is published only to month precision.

Benchmark scores

BenchmarkScore
A-Fantasia (Valid Attempts with Retries)72.3
A-Fantasia Chess (Valid Attempts with Retries)68.0
A-Fantasia Cube (Valid Attempts with Retries)79.0
A-Fantasia Spelling (Valid Attempts with Retries)70.0
AA-3D v1.0 Elo998
AA-AnalystAgent (pass@1)68.3
AA-AnalystAgent (pass@5)83.8
AA-AnalystAgent (pass^5)53.8
AA-Briefcase v1.1 Elo1575
AA-LCR v1.178.3
AA-Omniscience Index10.7
AA-Omniscience v1.1 Accuracy34.7
AA-Omniscience v1.1 Hallucination Rate34.2
AA-Omniscience v1.1 Index12.3
AdvancedIF80.8
AdvancedIF (rubric mean score)94.0
AIME 2025100.0
APEX (Mercor)15.2
APEX-Agents 1.168.3
APEX-Agents 1.1 Pass@156.1
APEX-SWE55.1
APEX-v1-extended64.7
ARC-AGI-1 Semi-Private89.0
ARC-AGI-261.3
ARC-AGI-3 (Semi-Private)2.8
Arena Code WebDev — Brand & Marketing1557
Arena Code WebDev — Consumer Product1516
Arena Code WebDev — Content Creation Tools1568
Arena Code WebDev — Data & Analytics1507
Arena Code WebDev — Frontend1604
Loading Atlas data…