Anthropic

Claude Opus 4.8 shipped as the new flagship model, with stronger coding, agentic, and knowledge-work performance.¹

Opus 4.8 defaults to a 1M token context window. 1M tokens by default on the Claude API, Amazon Bedrock, and Vertex AI, with 128k max output.² Mid-conversation system messages let a role: "system" turn follow a user message, preserving prompt-cache hits when instructions change; adaptive thinking reasons only when needed.³

Claude Code shipped orchestration and session management.

• The Workflow tool added deterministic multi-agent orchestration; dynamic workflows now fan work across hundreds of background agents⁴⁵
• Agent view (claude agents) tracks every running, blocked, or finished session, and /goal runs until a completion condition is met⁶⁷
• /simplify became /code-review for effort-leveled review with inline PR comments, and /usage breaks cost down by skill, subagent, plugin, and MCP server⁸⁹

Platform and API.

• Claude Platform on AWS exposes the Messages, Files, and Batches APIs plus Managed Agents with AWS billing and IAM auth; Claude in Amazon Bedrock opened to all customers¹⁰¹¹
• Cache diagnostics (beta) reports a cache_miss_reason when a prompt-cache prefix diverges, the Rate Limits API lets admins query configured limits, and web search now returns richer SEC filing data¹²¹³¹⁴

Managed agents, products, and governance.

• Multiagent sessions, Outcomes, and agent memory reached public beta; MCP tunnels (Research Preview) reach private-network servers and self-hosted sandboxes replace Anthropic-run tool execution^15161718
• Claude Cowork reached GA on macOS and Windows, Claude Design launched under Anthropic Labs, and Compliance API integrations plus SCIM-based role controls landed for enterprise governance^19202122

Deprecations. Claude Haiku 3 is retired.²³ Claude Sonnet 4 and Opus 4 retire June 15, 2026 — migrate to Sonnet 4.6 and Opus 4.8.²⁴ The 1M context beta for Sonnet 4.5 and Sonnet 4 is retired; use Sonnet 4.6 or Opus 4.6 instead.²⁵