AnthropicPro TierZero Data Retention

Sonnet 5

Anthropic's most agentic Sonnet yet (released Jun 30, 2026; knowledge cutoff Jan 2026).

Run Sonnet 5

Included on PRO plan

Context Window

Input Token Price

$2.20

per 1,000,000 tokens

Output Token Price

$11.00

per 1,000,000 tokens

Speed Rating

Moderate

Moderate (~94 tok/s)

Verified Frontier Benchmarks

Empirical Evaluation Results

Evaluated across standard and verified harnesses

SWE-bench Pro

coding

53.8%

Terminal-Bench 2.1

coding

80.5%

BrowseComp

agentic

84.7% single / 86.6% multi-agent (Sonnet 4.6: 76.2%)

HLE (no tools)

reasoning

41.3%

HLE (with tools)

reasoning

57.4%

OSWorld-Verified

agentic

81.2% (Sonnet 4.6: 78.5%)

FrontierCode v1

coding

38.8% (Sonnet 4.6: 15.1%)

AutomationBench

agentic

36.5%

HealthBench Professional

healthcare

57.8%

AA Intelligence Index

reasoning

38.2

AA Coding Index

coding

71.5

AnalystAgent

strategy

46.3%

AA Briefcase v1.1

strategy

1355 Elo

EnterpriseOps Gym

agentic

44.7%

GDPval-AA v2.1

strategy

1501 Elo

GPQA Diamond

reasoning

91.1%

Harvey LAB-AA

legal

90.1%

SciCode

science

54.3%

TAU-3 Banking

agentic

37.3%

Terminal-Bench 4.0

coding

14.1%

Vals EMB 2.0

finance

66.3%

Vals Vibe Code Bench

coding

81.3%

Vals Code Migration

coding

44.1%

Vals Legal Research

legal

41.8%

FrontierCode 1.1

coding

42.4%

FrontierMath Tier 4

math

65.6%

SimpleQA Verified

voice

33.7%

Toolathlon Verified

agentic

71.6%

CursorBench 4.0

coding

34.1%

Vals Finance Agent

finance

53.9%

Critpt

reasoning

16.9%

GDP PDF

strategy

13.2%

Lcr (1.1)

reasoning

82.0%

Mlcr

context

55.0%

MMMU Pro

reasoning

77.3%

Omniscience

science

16.5

ApprenticeBench CUA

agentic

16.0%

Apprenticebench API

reasoning

26.0%

Bullshitbench V1

reasoning

61.8%

Bullshitbench V2

reasoning

78.0%

Hyper Tau Bench (release-v1)

agentic

14.9%

Chess Puzzles

reasoning

16.0%

Mystery Game Puzzles

reasoning

35.0%

Context Arena MRCR V2 (8-needle)

context

80.0%

Mls Bench Lite (30-tasks)

reasoning

31.4

Surge Chartography (100-tasks)

reasoning

16.6%

Vals Index (2)

reasoning

59.6%

DeepSWE v1.1

coding

53.8%

AA-LCR v1.1

reasoning

82.0%

AA-Omniscience Index

reasoning

16.5

Frontiercode Extended (1.1)

coding

54.9%

Architectural Profile & Capabilities

Anthropic's most agentic Sonnet yet (released Jun 30, 2026; knowledge cutoff Jan 2026). **Closes much of the gap to Opus 4.8 at Sonnet pricing.** Plans, drives browsers and terminals, and runs autonomously at a level that previously required Opus. Near-Opus on HLE with tools (57.4% vs 57.9%). Configurable effort levels. 1M-token context window. New tokenizer (up to 1.35× more tokens than Sonnet 4.6 for same text). **Best for:** agentic coding, complex multi-step tasks, autonomous tool use, browser/terminal work, professional workflows, business automation. Introductory pricing $2/$10 through Aug 31, 2026; standard $3/$15 thereafter. **Caveat: at xhigh effort, cost can exceed Opus 4.8 at similar quality.**

Architecture Type: Dense Transformer

Standard dense autoregressive transformer architecture with full-attention mechanisms.

Recommended Workloads & Primary Use Cases

agentic coding

complex multi-step tasks

autonomous tool use

browser/terminal work

professional workflows

business automation. Introductory pricing $2/$10 through Aug 31

2026

standard $3/$15 thereafter

ARMES Zero Data Retention

Calls to Sonnet 5 are routed through strict Zero Data Retention inference channels. Prompts and outputs are never stored, indexed, or monitored by Anthropic.

View Privacy Architecture

More Models from Anthropic

Compare other engines available in this laboratory.

View Roster (12)

Opus 5.5Ultra Tier

Claude Opus 5.5 (anthropic/claude-opus-5.5) is Anthropic's September 22, 2026 flagship Opus-class reasoning model, succeeding Claude Opus 5 with a 1M-token context window, native image input, adaptive always-on extended thinking, and $4/$20 per million input/output tokens. It leads the frontier on agentic coding, terminal/computer-use, knowledge-work Elo (GDPval-AA v2.1: 1846), and long-horizon tool-use evals, at a reported ~97 tokens/sec on Benchmark Heaven telemetry.

1M ctx$4.00/M

Opus 5Ultra Tier

Anthropic's flagship 1M-context frontier multimodal foundation model, engineered for deep reasoning, autonomous coding agents, and complex long-horizon system execution.

1M ctx$5.00/M

Opus 4.8Ultra Tier

Anthropic's most capable model (released May 28, 2026; knowledge cutoff Jan 2026).

1M ctx$5.00/M

Architectural Profile & Capabilities

Architecture Type: Dense Transformer

Standard dense autoregressive transformer architecture with full-attention mechanisms.