Breaking

Anthropic Launches Claude Sonnet 5 With 1M-Token Context

June 30, 2026 at 20:40 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic on June 30, 2026 announced and released Claude Sonnet 5, the latest model in its Sonnet family. The company positions it as its "most agentic Sonnet yet," citing substantially strengthened planning, tool use across browsers and terminals, and autonomous task execution.

June 30, 2026 · Anthropic

Claude Sonnet 5: Opus-Class Agents at a Sonnet Price

Anthropic's "most agentic Sonnet yet" handles multi-step planning, tool use and self-verification in a single model — nearing top-tier Opus 4.8 performance while sitting in the mid-tier price bracket.

63.2%

Agentic coding score (up from 58.1%)

$2 / $10

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

~1–2M

Context window (tokens)

Agentic coding: Sonnet 5 lands between the two tiers

Column height ∝ benchmark score (%)

58.1%

Sonnet 4.6

63.2%

Sonnet 5

69.2%

Opus 4.8

The agent loop it now runs solo

Plan

→

Execute
browser · terminal · tools

→

Self-verify
without prompting

What's working

Finishes complex tasks earlier Sonnets abandoned midway
Traces root causes in existing code for durable fixes
End-to-end automation across CRM, notifications, legal and insurance workflows
Opus-class results with adjusted effort on BrowseComp & OSWorld-Verified

Where it lags

Mixed views on conversational naturalness
Questioned cost-effectiveness at high effort levels
0% success on a Firefox 147 exploit safety eval — well below Opus 4.8
Opus 4.8 still preferred on the hardest tasks

Rollout & pricing

Default model for Free & Pro on claude.ai from day one · available via Claude Code and API (claude-sonnet-5) · shipping on AWS Bedrock, Google Vertex AI and Microsoft Foundry. Intro price $2 / $10 per M tokens reverts to standard $3 / $15 after Aug 31, 2026.

Continue reading

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