Claude Tag brings collaborative AI agents to Slack as GPT-5.6 previews

Claude Tag: a new model for team AI agents

The biggest story this week is Anthropic's launch of Claude Tag, a new product that deploys one Claude instance per Slack channel. Instead of borrowing a user's identity, Claude Tag operates under its own scoped identity — a fundamental design choice that makes autonomous, multiplayer sessions safe and auditable. Teams can @Claude in any thread, and the agent picks up channel context, uses org tools, and persists across days. The agent identity model includes admin governance, token spend limits, and full audit logging. Available in beta for Claude Enterprise and Team customers, this represents Anthropic's most ambitious push into collaborative AI so far.

GPT-5.6 previews and the end of an era

OpenAI began a limited preview of the GPT-5.6 series: Sol (flagship), Terra (balanced), and Luna (fast and affordable). Sol launches with a new max reasoning effort and an ultra mode using subagents for complex work, paired with OpenAI's most robust safety stack including trained refusals, real-time misuse classifiers, and account-level review. Terra is competitive with GPT-5.5 while being 2x cheaper. Meanwhile, GPT-4.5 was retired from ChatGPT on June 25, with existing conversations migrating to GPT-5.5, and OpenAI o3 will follow on August 26 after a 90-day sunset period.

Codex Remote goes GA, Claude Code sharpens

Codex Remote reached general availability across all ChatGPT plans, using authenticated one-to-one QR pairing between mobile devices and Mac or Windows hosts. A new DigitalOcean Droplet Workspace plugin lets Codex provision a Droplet and connect it as a remote workspace. The feature is also available in ChatGPT Business and Enterprise/Edu workspaces.

On the Claude Code side, this week brought a steady stream of improvements. v2.1.186 added MCP server CLI auth (claude mcp login) and auto-responds to bash ! commands. v2.1.187 introduced sandbox credentials protection and org model restrictions. v2.1.191 fixed agent resurrection, streaming scroll jumps, and cut CPU usage by ~37%. v2.1.193 added OTEL assistant response logging and auto-mode shell routing, while v2.1.195 fixed voice dictation silence capture and hyphenated hook matchers. Anthropic also added Trusted Devices for remote control of Claude Code sessions.

Platform updates across the board

OpenAI's ChatGPT received several upgrades: improved dictation with a new speech-to-text model reducing word error rate by at least 10%, personal finance tools rolling out to Plus and Android users, and enhanced memory with source attribution across Business, Enterprise, and Edu plans. The model picker was simplified with new tiers replacing Thinking Standard/Extended/Heavy with Medium/High/Extra High. Plugin management controls arrived for Business and Enterprise/Edu workspaces.

On the API side, rate limits were raised across the Claude API, with Sonnet and Haiku now matching Opus at every tier, and usage tiers consolidated into three. OpenAI's Python SDK v2.44.0 fixed authorization header prioritization, while the Node SDK v6.45.0 added an afterCompletion hook to runTools and support for sideband call_id connections in realtime. Codex 0.142.2 improved MCP tool discovery and added proxy support for macOS and Windows.