Breaking

Anthropic Ships Claude Sonnet 5, Near-Opus 4.8 Performance at Lower Cost

June 30, 2026 at 16:45 EDT

Foundation Models
AI Agents

On June 30, 2026, Anthropic officially released Claude Sonnet 5, a new model with sharply improved agentic capabilities. It is positioned as a lower-cost option that delivers performance close to Claude Opus 4.8, the higher-tier model launched on May 28.

June 30, 2026 · Anthropic

Claude Sonnet 5: Opus-class agents, at the Sonnet price

Anthropic's new model sharpens autonomous tool use, planning and execution — aiming to deliver performance close to the higher-tier Opus 4.8 for a fraction of the cost.

$2 / $10

Intro price per 1M tokens (in/out), through Aug 31, 2026

Token context window, up to 128k output tokens

Platforms at launch: claude.ai, Code, API, Bedrock, Vertex AI

Output price per 1M tokens — standard rates

Sonnet 5 lands well below the higher tiers it competes with on capability.

$15

Sonnet 5

balanced

$25

Opus 4.8

high performance

$50

Fable / Mythos 5

frontier

Each block ≈ $5 output / 1M tokens. Sonnet 5 introductory output rate is just $10.

What partners praise

Higher completion on complex tasks
Self-verification cuts follow-ups
Strong on existing (brownfield) code
Token-efficient, fewer turns
Useful in legal, finance, insurance

What critics flag

Opus 4.8 can be cheaper per task
Higher token use may offset savings
More hallucination & flattery reported
Value proposition called unclear

Cybersecurity guardrails

Sonnet 5's cyber capability is intentionally held far below Opus 4.8 and the Mythos tier.

Fable 5 falls back to Opus 4.8 for cyber, biology and chemistry queries. Mythos 5 stays restricted via Project Glasswing — a limited U.S. government and cyber-defender program — with no public release date set while stronger safeguards are built.

Where Sonnet 5 is headed

Multi-step software engineering → Agentic coding → Data pipelines → Legal & finance → Computer-use workflows

How far it eases the price–performance trade-off will shape its assessment as AI agents reach practical deployment.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.